Mas e Thesis on Sound and Music Compu ing
Uni e si a Pompeu Fab a
MuSA: A New TEL Pla o m o
Enhancing Sel -Re lec ion and Musical
Unde s anding h ough Saliency Analysis
o Pe o mance Reco dings
Isabelle Ok ay
Supe iso : Ra ael Rami ez-Melendez
Co-Supe iso : Su i Haeae ae
July 2025
Con en s
1 In oduc ion 1
1.1 Mo i a ion.................................. 2
1.2 Objec i es.................................. 5
1.3 CodeAccessibili y ............................. 7
2 Backg ound 8
2.1 Founda ions o Musical Talen De elopmen . . . . . . . . . . . . . . . 8
2.1.1 TheTADMusicModel........................... 9
2.2 P ac ice S a egies in Music Educa ion . . . . . . . . . . . . . . . . . . 12
2.2.1 The Mas e -App en ice Model . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Cons uc i ism and Dialogic Teaching . . . . . . . . . . . . . . . . . . 13
2.2.3 Sca olding.................................. 14
2.2.4 Sel -Regula ed Lea ning . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.5 Sel -Di ec ed Lea ning . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Feedback................................... 19
2.3.1 Cogni i eLoad ............................... 20
2.3.2 Modeling .................................. 22
2.3.3 Salien Momen s .............................. 23
2.3.4 Syn hesizing SRL, SDL, Feedback, and MuSA . . . . . . . . . . . . . . 23
2.4 Technological Pedagogical Con en Knowledge . . . . . . . . . . . . . . 24
2.5 TEL Feedback Sys ems in Music Educa ion . . . . . . . . . . . . . . . 25
2.5.1 OnlineResou ces .............................. 27
2.5.2 Non-Real-Time Reco ding Feedback . . . . . . . . . . . . . . . . . . . . 29
2.5.3 Real-Time Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6 Analyzing Salien Pe o mance Momen s . . . . . . . . . . . . . . . . . 32
2.7 Backg ound Resea ch Summa y . . . . . . . . . . . . . . . . . . . . . . 33
3 Designing and Implemen ing MuSA 34
3.1 Ini ialP o o ype .............................. 34
3.2 Implemen ing MuSA-V2 . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Selec ing Musical Fea u es o MuSA-V2 . . . . . . . . . . . . . . . . . 38
3.2.2 DesigningMuSA .............................. 42
3.3 MuSA Sys em A chi ec u e . . . . . . . . . . . . . . . . . . . . . . . . 43
4 MuSA E alua ion S udy 49
4.1 Me hods................................... 50
4.2 Resul s.................................... 53
4.2.1 Objec i e Pe o mance Me ics . . . . . . . . . . . . . . . . . . . . . . 54
4.2.2 Pe cei ed Pe o mance Me ics . . . . . . . . . . . . . . . . . . . . . . 59
4.2.3 Sel -Awa eness Me ics . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2.4 Sen imen Analysis............................. 65
5 Discussion 69
5.1 Discussion.................................. 69
5.1.1 In eg a ing Theo y in o P ac ice . . . . . . . . . . . . . . . . . . . . . . 69
5.1.2 E alua ing MuSA: Demand, Limi a ions, and Ou comes . . . . . . . . 71
5.1.3 Independen P ac ice and Lea ne Implica ions . . . . . . . . . . . . . 74
5.1.4 MuSA’s Limi a ions and Fu u e Di ec ions . . . . . . . . . . . . . . . . 75
5.2 Conclusions ................................. 80
Lis o Figu es 82
Lis o Tables 86
Bibliog aphy 88
A Appendix: MuSA E alua ion S udy Resul s 114
Acknowledgemen
This p ojec would no ha e been possible wi hou he guidance and suppo o my
supe iso , Ra ael Rami ez-Melendez. He in oduced me o SkyNo e and allowed me
o wo k on upda ing he SkyNo e web applica ion, an o shoo o he o iginal TELMI
[1] p ojec . I enjoyed wo king on SkyNo e so much ha when Ra ael p oposed he
p ojec on saliency analysis and c ea ing a non- eal- ime music pe o mance eedback
applica ion, I was h illed and immedia ely chose i as my hesis opic. Thank you,
Ra ael, o all you guidance and insigh h oughou his p ojec —I ha e lea ned
immensely, and I hope ha someone will con inue he nex i e a ion o MuSA.
I am also deeply g a e ul o my co-supe iso , Su i Hää ä, who s udied alongside me
du ing he Sound and Music Compu ing mas e ’s p og am a Uni e si a Pompeu
Fab a (2023–2024). Su i has suppo ed me on nume ous p ojec s, including his
hesis, and has consis en ly been a posi i e in luence, helping me s ay ocused and
mo i a ed. Thank you, Su i, o all you encou agemen and ha d wo k—you
suppo has been in aluable.
Thank you o Anmol Mish a, who assis ed me du ing a c i ical momen in deploying
he applica ion. Ha ing ne e deployed any hing be o e, I was qui e los , bu Anmol
pa ien ly guided me h ough mul iple calls, e en while in India, o help sol e he
p oblem. You help was essen ial— hank you, Anmol!
I also owe immense g a i ude o my mo he , who was always he e when I doub ed
mysel o el o e whelmed by his p ojec . He guidance, encou agemen , and
wisdom, pa icula ly as a physician and academic, kep me mo i a ed and g ounded
h oughou his jou ney.
Finally, hank you o my pa ne , Théo Fuh mann, o unwa e ing suppo , o
cooking meals du ing ma a hon coding sessions, and o helping me main ain pe -
spec i e du ing long days o wo k. You companionship, pa ience, and willingness
o oubleshoo alongside me, some imes as my ubbe duck, kep my spi i s high
and made his p ojec possible.
Abs ac
This hesis explo es how a echnology-enhanced lea ning (TEL) ool can imp o e
music p ac ice by add essing a c i ical, o en-neglec ed componen o skill de elop-
men : he e lec ion phase. I ocuses on he de elopmen and e alua ion o MuSA
(Musical Salience Analyze ), an applica ion designed o p o ide a pedagogically-
g ounded pla o m o analyzing eco ded pe o mances o make e lec ion mo e
e icien and e ec i e. MuSA’s design is in o med by key educa ional heo ies, in-
cluding he Talen -De elopmen -in-Achie emen -Domains (TAD) Music Model and
lea ne -cen e ed eaching (LCT) p inciples like sca olding, sel - egula ed lea ning
(SRL), and sel -di ec ed lea ning (SDL). I s cen al ea u e is saliency analysis,
which algo i hmically iden i ies key momen s in a pe o mance based on a iabil-
i y in musical ea u es such as pi ch, dynamics, and empo. Unlike ools ha o e
p esc ip i e, "co ec /inco ec " eedback, MuSA encou ages a lea ne ’s own in e -
p e a ion. As an accessible, web-based pla o m, i allows use s o upload o eco d
audio o analysis independen ly o a eache . To e alua e MuSA’s e ec i eness, a
mixed-me hods, wi hin-subjec s s udy was conduc ed wi h 14 pa icipan s. While
he s udy’s small sample size limi ed s a is ical powe , he indings poin ed o se e al
explo a o y ends. The da a sugges ed a di e en ial impac based on musical ea-
u e and expe ience le el, wi h dynamics showing he mos consis en end owa d
objec i e and pe cei ed imp o emen . The analysis also sugges ed a po en ial ex-
pe ise e e sal e ec , whe e ends showed in e media e musicians gaining om he
eedback while ad anced musicians expe ienced neu al o sligh ly nega i e changes.
Fu he mo e, he s udy’s sel -awa eness me ics indica ed a gene al misalignmen
be ween pa icipan s’ sel - a ings and objec i e pe o mance, highligh ing a co e
challenge in he sel - e lec ion phase o independen p ac ice. In conclusion, MuSA
o e s a po en ial con ibu ion o TEL o music by le e aging compu a ional anal-
ysis o p o ide a ge ed insigh s ha can sca old he e lec i e p ocess. Al hough
he quan i a i e esul s we e inconclusi e, posi i e quali a i e eedback alida es he
demand o such a ool. This wo k p o ides a unc ional p o o ype and a esea ch
in as uc u e o collec ing labeled eco ding da a, demons a ing he dual ole o
6Chap e 1. In oduc ion
o musical alen de elopmen and how s uden s p og ess h ough hem. Un-
de s anding hese s ages allows o a heo e ical e alua ion o how MuSA can
suppo lea ne s as hey de elop musical alen .
2. Compile a backg ound o common p ac ice s a egies and lea ning
amewo ks used in bo h independen and adi ional mas e -app en ice se -
ings. Doing so in o ms how MuSA can be designed o in eg a e smoo hly in o
exis ing p ac ice ou ines and acili a e pe o mance e lec ion.
3. Emphasize musical exp ession as a co e dimension o musical alen
o guide MuSA’s de elopmen owa d p o iding lexible, pe sonalized eedback
( os e ing exp essi e skills) a he han solely bina y “co ec ”/“inco ec ” judg-
men s ( aining echnical skills).
4. E alua e exis ing TEL pla o ms o music p ac ice and pe o mance
analysis, ocusing on hei capaci y o pe sonalized eedback and suppo o
lea ning musical exp ession. This includes e iewing online lea ning s a egies
and assessing bo h eal- ime and non- eal- ime eedback applica ions (espe-
cially pe o mance analysis ools) o iden i y gaps in TEL music pla o ms.
To e alua e he pedagogical alue o MuSA, a mixed-me hods s udy was conduc ed.
The s udy examined whe he de ec ing salien momen s and p o iding isual eed-
back a ec ed musicians’ sel - a ed and objec i e pe o mance (i.e., e lec ion) on
ea u es like pi ch, dynamics, and empo. The key con ibu ions o his hesis in-
clude:
1. The design and implemen a ion o MuSA, a TEL pla o m ha analyzes
salien momen s o a music pe o mance based on di e en musical aspec s
and can be used ou side he adi ional mas e -app en ice model o acili a e
pe o mance e lec ion.
2. Empi ical indings om a mixed-me hods use s udy ha may in o m
u u e in eg a ions o pe o mance analysis echnologies in o independen mu-
sic lea ning, speci ically wi h ega ds o sel -awa eness du ing music pe o -
mance.
3. A di ec ion o u u e explo a ions ha can u he de elop he MuSA
1.3. Code Accessibili y 7
web applica ion and explo e how p esen ing salien momen s o lea ne s may
no only suppo pe o mance co ec ion ela i e o a e e ence audio bu also
enhance musical unde s anding and e lec ion du ing p ac ice h ough he i-
sualiza ion o pe o mance eco dings.
The hesis is s uc u ed as ollows: Chap e 1 in oduces he esea ch con ex and
mo i a ion; Chap e 2 e iews ele an li e a u e and he s a e o he a ; Chap e
3 de ails he design and de elopmen o MuSA; Chap e 4 ou lines he e alua ion
s udy done on MuSA; and Chap e 5 discusses implica ions and u u e wo k.
1.3 Code Accessibili y
All code is accessible h ough Gi hub6. Bo h MuSA7and he MuSA e alua ion
s udy8can be accessed online.
6h ps://gi hub.com/isabelleok ay/audio-analyze /
7h ps://analyze .appskyno e.com/
8h ps://analyze .appskyno e.com/ es ing/
Chap e 2
Backg ound
2.1 Founda ions o Musical Talen De elopmen
Musical alen is o en pe cei ed as inna e, ye i mo e accu a ely eme ges om he
in e ac ion be ween indi idual p edisposi ions and cul i a ed skills [3]. Recognizing
his in e play is essen ial o explaining why some musicians excel and o iden i ying
how echnology-enhanced lea ning (TEL) can bes suppo hei de elopmen . This
pe spec i e no only jus i ies he c ea ion o MuSA bu also cla i ies he pedagogical
p inciples ha shape i s co e unc ionali y.
Some elemen s o musical alen a e less easily in luenced. In elligence, o example,
is la gely de e mined by gene ics [69] and comp ises a ange o cogni i e abili ies
such as pe cep ual speed [70] and easoning abili y [71]. Indi iduals also di e in
hei audi o y senso y disc imina ion e en be o e ecei ing o mal musical educa ion,
a capabili y known as musical ap i ude [3]. En i onmen al ac o s can also
p edispose alen , as child en whose abili ies a e ecognized ea ly by pa en s a e
mo e likely o ecei e en iched musical oppo uni ies [72]. O he aspec s a e mo e
di ec ly shaped h ough in en ional e o . P ac ice, o ins ance, is widely ega ded
as a pilla o alen de elopmen , as no one is bo n knowing how o ead shee music,
play scales, o sigh -sing. Ea ly engagemen wi h music u he de elops echnical
skills and os e s pe sonali y ai s such as willpowe and mo i a ion, c ea ing a
8
2.1. Founda ions o Musical Talen De elopmen 9
ounda ion o he con inual acc ual o musical knowledge [73, 74].
Howe e , he ela i e impac s o hese ac o s on de eloping musical alen emain
deba ed, wi h p ac ice being a ocal poin o con en ion. In his discussion, he e m
p ac ice e e s speci ically o delibe a e p ac ice, which is unde s ood as a se o
ac i i ies ypically designed by an ex e nal sou ce, o en a eache , wi h he explici
goal o imp o ing pe o mance [6, 2, 7, 75]. This de ini ion p o ides a na owe ,
mo e ope a ional measu e o p ac ice, aligning wi h much o he li e a u e agains
which compe ing in luences (e.g., in elligence, musical ap i ude) ha e been e alua ed
[4, 5].
Resea ch shows ha accumula ed p ac ice hou s a e he s onges p edic o o ech-
nical skills like sigh - eading [6, 7]. Ye me a-analyses es ima e p ac ice explains
only 23% o 37% o he a iance in o e all music achie emen [4, 5]. C i ics o
hese indings ci e me hodological issues, including inconsis en de ini ions o de-
libe a e p ac ice [76]. E en so, gene ic ac o s such as in elligence may su pass
p ac ice in p edic ing musical de elopmen —especially in beginne s—and may p e-
dic sigh - eading pe o mance a e con olling o p ac ice [77, 78, 7]. En i on-
men al in luences also in e ac wi h gene ic ones, as musically inclined pa en s may
bo h os e en iched en i onmen s and pass on ai s p edisposing child en o o -
mal aining. Such inhe i ed endencies o selec and shape en i onmen s, known
as gene–en i onmen ansac ions [69], sugges ha musical expe iences may
hemsel es be pa ly he i able.
2.1.1 The TAD Music Model
While p ac ice is i al o de eloping ins umen al echnique and pe sonal skills, i
ope a es alongside—and in cons an in e ac ion wi h—gene ic, cogni i e, and en i-
onmen al in luences. To si ua e hese ac o s wi hin a cohe en de elopmen al a-
jec o y, his hesis adop s he Talen -De elopmen -in-Achie emen -Domains
(TAD) Music Model [44, 45]. This empi ically alida ed amewo k ou lines dis-
inc s ages o alen acquisi ion and iden i ies psychological p edic o s ha ack
p og ession o e ime, in eg a ing bo h p edisposi ions and lea ned skills. Using
10 Chap e 2. Backg ound
his model allows MuSA’s design o be ailo ed owa d a clea ly de ined audience o
lea ne s.
The TAD Music Model is no he i s a emp o concep ualize musical alen
de elopmen , bu i is dis inc i e in i s use o measu able p edic o s o link ime
and achie emen [79]. Ea lie models o e emphasized gene ic and en i onmen al
ac o s [80, 81, 82, 83], which a e challenging o e alua e empi ically due o he
cos , complexi y, and sca ci y o longi udinal s udies [84, 42, 85]. As a esul , pu ely
gene ic- o en i onmen -based models ha e p o en less eliable o p edic ing long-
e m musical achie emen .
Figu e 1: The Talen -De elopmen -in-Achie emen -Domains F amewo k [44]
The TAD Music Model (Figu e 1) iden i ies ou s ages o musical alen de elop-
men , each wi h cha ac e is ic p edic o s and lea ning needs:
1. Ap i ude: inna e men al and physical abili ies, such as melodic, hy hmic,
and onal sensi i i y, ha p edic a na u al capaci y o musical success [86].
2. Compe ence: he acquisi ion o skills needed o independen p ac ice, ex-
p essi e imp o emen , and epe oi e expansion. Key p edic o s include a
2.1. Founda ions o Musical Talen De elopmen 11
g ow h mindse [87, 88], sel -mo i a ion [79, 89], musical sel -concep [89, 88],
mo o skills [90], and delibe a e p ac ice.
3. Expe ise: a le el o achie emen cha ac e ized by pee ecogni ion, c ea i e
p oblem-sol ing, and public engagemen [91]. P edic o s include psychological
s abili y, openness, conscien iousness, esilience, and ocused dedica ion [79].
4. T ans o ma ional achie emen : a le el o expe ise ma ked by signi ican ,
in luen ial, and c ea i e con ibu ions. This s age is less p edic able, shaped
by ac o s such as chance, oppo uni y, c ea i e po en ial [92], musical o m
[93, 94], and s a egies o sus aining high-le el ou pu [95].
The TAD Music Model shows ha musicians a di e en s ages need di e en kinds
o suppo . Fo example, in he Ap i ude s age, a playe needs basic echnical ain-
ing o u n na u al abili y in o co e echnical skills. These include sigh - eading,
pos u e, ea - aining, and he abili y o play a piece wi h ega ds o musical aspec s
like iming, in ensi y, pi ch, and imb e. On he o he hand, in he Compe ence
s age, he ocus shi s o s eng hening echnique while also de eloping a pe sonal
s yle and exp essi e oice. Musical exp ession in ol es shaping musical aspec s
o con ey a wo k’s emo ional cha ac e and connec a pe o me ’s in e p e a ion
wi h a lis ene ’s expe ience [43, 96].
As Sec ion 2.5 demons a es, many TEL music ools ocus on echnical skills, p o-
iding bina y eedback (co ec /inco ec ) ied o he musical sco e. This app oach
misses he in e p e i e dialogue essen ial o mode n eaching amewo ks and lea es
a gap o musicians who need o de elop hei exp essi e abili ies. Mo eo e , such
ools a e necessa y o guide he e lec i e aspec o p ac ice, which is guided by he
lea ne ’s in e p e a ion o hei pe o mance a e ac i ely p ac icing. The e o e, in
addi ion o echnical skills, we de elop MuSA in such a way as o acili a e musical
exp ession in pa allel, as bo h a e essen ial o de eloping musical alen and a e a
co e ocus o TEL in music educa ion.
12 Chap e 2. Backg ound
2.2 P ac ice S a egies in Music Educa ion
Building on he ac o s shaping musical alen discussed in Sec ion 2.1, his hesis
ocuses on p ac ice, since i is he ac o ha lea ne s can con ol mos . To design
TEL p ac ice ools e ec i ely, i is impo an o unde s and he di e en ypes o
p ac ice ha in luence alen de elopmen . E icsson (2016) iden i ied h ee ypes
[97]:
1. Nai e p ac ice: Playing an ins umen o enjoymen wi hou speci ic goals
o s uc u ed imp o emen .
2. Pu pose ul p ac ice: Goal-o ien ed p ac ice done independen ly, wi hou
con inuous expe guidance o s uc u ed eedback.
3. Delibe a e p ac ice: Highly s uc u ed p ac ice guided by a eache wi h
deep expe ise.
The di e ence in pe o mance ou comes be ween pu pose ul and delibe a e p ac ice
is subs an ial. In E icsson (1993) seminal s udy o iolinis s a he Music Academy
in Wes Be lin, he mos accomplished musicians had accumula ed he highes num-
be o delibe a e p ac ice hou s [6]. La e me a-analyses sugges ed ha delibe a e
p ac ice accoun ed o only 23% [4] o 37% [5] o pe o mance a ia ion. E icsson e
al. (2019) a gued ha hese s udies used inco ec de ini ions, and when delibe a e
p ac ice is p ope ly de ined, i s explana o y powe ises o 61% [76], indica ing a
s ong in luence in musical alen de elopmen . TEL ools like MuSA suppo de-
libe a e p ac ice by s eng hening he de elopmen o musical alen while making
lea ning mo e e ec i e and accessible h ough cogni i e and pedagogical s a egies.
2.2.1 The Mas e -App en ice Model
Gi en he cen al ole o delibe a e p ac ice, i is unsu p ising ha adi ional music
educa ion has long elied on he mas e –app en ice model, one popula lea ning
s a egy. In his app oach, s uden s ecei e in-pe son guidance om an expe ienced
u o , and he musical sco e se es as he co e o ins uc ion [98, 99, 100, 101,
10, 11]. Despi e he ise o independen lea ning and online esou ces, his model
2.2. P ac ice S a egies in Music Educa ion 13
emains dominan in conse a o ies and highe educa ion wo ldwide [102], ac oss
bo h Wes e n and non-Wes e n adi ions [103].
Mos lea ne s do no ha e cons an access o de ailed eedback a ailable h ough
conse a o ies and mas e -app en iceship ha can enable e lec ion, ecei ing only
a single weekly lesson [104]. Le o p ac ice alone o mos o hei ime, hese
lea ne s may de elop inco ec echniques, lose di ec ion, o o m ine ec i e habi s.
Wi hou consis en guidance, e en mo i a ed s uden s can educe delibe a e p ac ice
o pu pose ul, o e en nai e, p ac ice [105, 9, 8, 76].
To add ess his challenge, expe s ha e de eloped eaching app oaches ha ocus on
he lea ne and hei needs a he han jus he u o o musical sco e. The ollowing
sec ions examine cons uc i is s a egies such as dialogic eaching, sca olding, and
modeling, which can p omo e lea ne au onomy, de elop skills o mo e e ec i e
independen p ac ice, and acili a e e lec ion. These amewo ks guided MuSA’s
design, p o iding a basis o suppo and enhance p ac ice when expe guidance is
limi ed.
2.2.2 Cons uc i ism and Dialogic Teaching
Cons uc i is lea ning p inciples posi ion he lea ne as an ac i e pa icipan a he
han a passi e ecipien o ins uc ion. In cons uc i is heo y, he eache
ac s as a acili a o , designing ac i i ies ha help lea ne s build unde s anding by
eo ganizing and in eg a ing p io knowledge [14, 106, 107, 15, 108]. This lea ne -
cen e ed eaching (LCT) , in con as o he mas e -app en ice model’s emphasis
on eache and sco e au ho i y, can p o ide he bene i s o expe eedback while
also os e ing a collabo a i e en i onmen ha gi es a s uden owne ship o e hei
lea ning [109]. Componen s o LCT include he use o in e ac i e, pe sonalized
lea ning ac i i ies and he inclusion o echnology o manage lea ning ou side o
he class oom. LCT has shown o lead o highe lea ning ou comes and mo i a ion
ac oss a a ie y o ields including medicine [110], ma hema ics [111], and music
[112, 113].
14 Chap e 2. Backg ound
A p ac ical exp ession o LCT in music educa ion is dialogic eaching, an ap-
p oach in which sus ained dialogue be ween eache and s uden becomes he p i-
ma y medium o lea ning. In dialogic eaching, a he han deli e ing one-way
ins uc ions, eache s engage lea ne s h ough ques ioning, e lec ion, and collabo-
a i e explo a ion, enabling eal- ime adap a ion o guidance o mee e ol ing needs
[114, 115]. This open eedback channel encou ages s uden s o co-cons uc hei
own in e p e a ions, which is why MuSA’s pe o mance eedback a oids bina y co -
ec /inco ec labeling: MuSA is designed as a lea ne -cen e ed ool o e lec ion
a he han a sco e-cen e ed ool o co ec ion.
2.2.3 Sca olding
Ano he LCT app oach is sca olding, which in ol es ailo ing suppo o a lea ne ’s
cu en abili y and g adually ans e ing esponsibili y o pe o mance and in e -
p e a ion o he s uden . This p ocess is oo ed in he wo k o Vygo sky, a co e
con ibu o o cons uc i ism. I le e ages his concep o he zone o p oximal
de elopmen , which is he ange o asks a lea ne can pe o m wi h assis ance
bu no independen ly, and p o ides ailo ed suppo ha g adually dec eases as
he lea ne ’s compe ence g ows [106, 116]. The key p inciples o sca olding (Fig-
u e 2) a e:
1. Con ingency: The eache ’s suppo mus be con inually adap ed o he
lea ne ’s cu en skill le el, p e en ing he lea ne om being o e whelmed o
unchallenged [117, 118, 119].
2. Fading: Suppo is g adually dec eased as he lea ne ’s compe ence g ows.
Fo example, a eache may ini ially p o ide ex ensi e modeling bu educe
assis ance as he lea ne becomes mo e p o icien wi h a skill o piece [120, 118].
3. T ans e o Responsibili y: The goal o sca olding is o shi lea ning e-
sponsibili y o he lea ne , os e ing au onomous compe ence so hey can pe -
o m asks independen ly [121].
4. Cyclical P ocess: Sca olding is a ecu ing cycle. Once a lea ne mas e s
one sub-goal (e.g., a basic bow hold on a iolin), he eache in oduces a new,
2.2. P ac ice S a egies in Music Educa ion 15
sligh ly mo e challenging sub-goal. The sca olding p ocess hen begins again
o his new ask [122].
Figu e 2: Concep ual model o sca olding [118].
MuSA can be con ex ualized wi hin he sca olding amewo k by p o iding a o m o
con ingency and acili a ing he ans e o esponsibili y by ac ing as an SRL ool.
While sca olding ypically elies on a dedica ed eache o ailo suppo , MuSA’s
eedback can ac as an TEL in e media y. Fo example, i a lea ne is ocused on
a speci ic pe o mance ou come, hey can analyze a pe o mance eco ding wi h
MuSA. The pla o m hen esponds by highligh ing in e es ing salien momen s (see
Sec ion 2.3.3) du ing di e en musical aspec s. This e lec s how a eache p o ides
con ingency by indica ing salien momen s in a s uden ’s pe o mance ha may
no be ob ious o hem. The goal is o ini ia e a ans e o esponsibili y o ha
speci ic ask by helping he lea ne unde s and wha o lis en o , wha pe o mance
ou comes o s i e o , and wha ools exis o help hem p ac ice.
22 Chap e 2. Backg ound
2.3.2 Modeling
One popula me hod o gene a ing eedback in music educa ion and po en ially
educing cogni i e load when lea ning a new ask is modeling, in which a eache
demons a es a desi ed sound o echnique o en based on a lea ne ’s cu en pe o -
mance s a e [139]. Modeling can ake wo o ms: au al modeling, whe e lea ne s
lis en o and emula e expe pe o mances o build in e nal musical ep esen a ions
[51, 140, 141, 142], and isual modeling, which emphasizes physical aspec s like
pos u e and inge ing wi h hea y eliance on w i en no a ion [143].
Au al e sus isual modeling. While bo h isual and au al modeling ha e
alue, esea ch shows ha au al modeling has highe lea ning ou comes o bo h
echnical skills and musical exp ession since au al modeling ocuses on ex e nal
ou comes (how he ins umen sounds) a he han jus he in e nal ac ion (how
he body pe o ms) [144, 145, 68, 146, 147, 148, 113]. This conclusion aligns wi h
Meissne ’s (2021) dialogic eaching amewo k, which combines cogni i e load he-
o y, dialogic eaching, and modeling-based eedback, sugges ing ha ocusing on he
o e all ou come a he han becoming en angled in minu ia can manage cogni i e
load [149]. Addi ionally, o e - eliance on isual modeling may weaken a s uden ’s
abili y o connec hei physical ac ions wi h he esul ing sound [150]. As a esul ,
musicians may s uggle o de elop musical unde s anding, o he abili y o ec-
ognize and manipula e pa e ns and ela ionships in music [151]. When modeling
emphasizes symbolic in o ma ion, i becomes sco e-cen e ed a he han lea ne -
cen e ed, ocusing on wha o do ins ead o how and why, hus ailing o wo k
owa ds musical unde s anding. This is one eason MuSA elies on he sound o an
indi idual’s pe o mance ins ead o he shee music behind i .
Mul i-Modal Modeling. Modeling is o en pai ed wi h o he lea ning me hods
in a combina ion known as mul i-modal modeling o u he enhance lea ning
ou comes. Fo ins ance, while modeling alone can os e dependence, combining i
wi h s a egies like e bal eedback encou ages s uden s’ in e p e i e hinking and
in e nal musical ep esen a ions [149]. This is because e bal eaching me hods,
2.3. Feedback 23
such as using me apho s o explici explana ions, a e shown o imp o e musical
unde s anding and s eng hen exp essi e pe o mance [152], which is in line wi h
LCT dialogic eaching ou comes [112, 113].
2.3.3 Salien Momen s
O en, when p o iding eedback using modeling, eache s na u ally highligh impo -
an pe o mance momen s by exagge a ing musical aspec s like in ona ion, dynam-
ics, o iming du ing model pe o mances [67]. This hesis de ines hese highligh ed
momen s as salien momen s. A eache may indica e salien momen s as a eas
du ing a pe o mance o echnical co ec ion o hose whe e a lea ne can employ
hei own musical in e p e a ion ela i e o an exis ing no m (like w i en no a-
ion). These may include pi ch e o s, iming issues, o exp essi e changes ha
dese e close a en ion. Especially when ocusing on musical exp ession, iden i y-
ing a salien momen does no inhe en ly imply assigning posi i e o nega i e alue
o i ; a he , he iden i ica ion o a salien momen can s a con e sa ion abou
exp essi e in en (dialogic eaching) o cla i y i a lea ne ’s echnique o choices a e
ac ually leading o hei in ended exp essi e ou come.
2.3.4 Syn hesizing SRL, SDL, Feedback, and MuSA
Looking a he TAD Model, delibe a e p ac ice, sca olding, and SRL, one heme
s ands ou : e lec ion is essen ial a all s ages o alen de elopmen o u ning
epe i ion in o eal p og ess, ye i is o en neglec ed. Lea ne s may plan and p ac-
ice ca e ully, bu wi hou s uc u ed oppo uni ies o e alua e hei pe o mance,
p ac ice can become mechanical and disconnec ed om long- e m goals [153]. As
Sec ion 2.5.2 discusses, many music lea ne s use sel - eco ding o e lec on hei
playing, bu his me hod is o en ine icien : e iewing ull sessions akes ime [154],
adds cogni i e load, and can lead o passi e lis ening ha misses sub le mis akes
[138]. As a esul , e lec ion ends o be occasional a he han sys ema ic. MuSA
add esses his p oblem by au oma ically highligh ing salien momen s and ocusing
e lec ion asks. By educing he ime and e o needed o e iew, a ool like MuSA
24 Chap e 2. Backg ound
can make e lec ion easie o do egula ly. This app oach helps in eg a e e lec ion
in o he p ac ice cycle, imp o es planning h ough eedback on ecu ing pa e ns
(e.g., “in ona ion d i s sha p in high posi ions”), and suppo s independen lea ning
while hope ully de eloping c i ical lis ening and sel -assessmen skills.
2.4 Technological Pedagogical Con en Knowledge
Designing a new TEL ool o suppo he e lec ion p ocess in music p ac ice, such as
MuSA, equi es guidance beyond b oad psychological o cogni i e p inciples. Models
such as Technological Pedagogical Con en Knowledge (TPCK) and i s ex-
pansion h ough Technology Mapping (TM) p o ide a s uc u ed way o connec
hese p inciples o he speci ic challenges o music educa ion. The ollowing sec ion
ou lines how his model, which emphasizes he in e play o echnology, pedagogy,
con en , and he a ec i e domain, in o ms and jus i ies MuSA’s design.
TPCK. A ound 20 yea s ago, esea che s in oduced Technological Pedagogical
Con en Knowledge (TPCK) o guide he design o TEL ools by emphasizing how
subjec knowledge, pedagogy, and echnology in e sec in eaching [155, 17, 18].
Angeli and Valanides (2013) expanded his amewo k wi h an ins uc ional design
app oach called Technology Mapping, which speci ies how ools should be de eloped
o use echnology’s unique a o dances o lea ning [19]. The p inciples o TPCK
include:
1. Find high-impac con en : Iden i y opics ha a e di icul o s uden s o
unde s and o o eache s o explain wi hou he aid o echnology.
2. C ea e unique ep esen a ions: De elop new ways o p esen con en ha
a e only possible wi h echnology, making i easie o g asp.
3. Le e age no el eaching me hods: Iden i y and use eaching s a egies
ha a e di icul o impossible o implemen wi h adi ional me hods alone.
4. Choose he igh ools: Selec echnology wi h he app op ia e ea u es o
suppo he eaching and lea ning goals.
5. Design lea ne -cen e ed ac i i ies: C ea e lea ne -cen e ed class oom ac-
2.5. TEL Feedback Sys ems in Music Educa ion 25
i i ies ha e ec i ely in eg a e echnology.
TM and he A ec i e Domain. Mac ides and Angeli (2018) expanded Tech-
nology Mapping o music educa ion by adding he a ec i e domain (Figu e 5)
[156, 157]. This addi ion highligh s ha emo ion, mo i a ion, and pe sonal ex-
p ession a e cen al o music lea ning, add essing a gap in he o iginal amewo k
[158], which only ocused on pedagogical and cogni i e aspec s. Resea ch shows
ha many s uden s s uggle wi h composi ion, imp o isa ion, and lis ening because
o limi ed skills o con idence [159, 160, 161]. By in eg a ing he a ec i e domain,
he expanded model e ames hese challenges as oppo uni ies o connec echnical
lea ning wi h exp ession and emo ion. Educa o s a gue ha musical expe iences
canno be sepa a ed om hei exp essi e cha ac e [162], and ha emphasizing
sel -exp ession inc eases engagemen and mo i a ion [8, 151].
Using TPCK o In o m MuSA. MuSA’s design e lec s his expanded model,
which s esses ha music lea ning in ol es bo h cogni i e and emo ional aspec s
[163]. MuSA p o ides lexible, non-co ec i e eedback ha s uden s can in e p e
in ways ha sui hei needs, echoing he iew ha ocusing on exp ession can boos
engagemen [8]. By highligh ing salien sec ions o a eco ding, MuSA di ec s a en-
ion o meaning ul changes in dynamics, empo, o a icula ion, ea u es known o
in luence emo ional esponses [158, 164]. This app oach lowe s ba ie s o lea ne s
wi h di e en skill le els, since he ocus is on eco ded analysis a he han li e
pe o mance [27]. A he same ime, i s saliency analysis suppo s SRL by helping
s uden s e lec on hei playing and ake mo e con ol o hei p og ess, in line wi h
TM’s emphasis on in e ac i e, lea ne -cen e ed use o echnology [19, 163].
2.5 TEL Feedback Sys ems in Music Educa ion
A wide ange o TEL ools can suppo music lea ning, wi h di e en ools add ess-
ing di e en TPCK p inciples depending on hei pu pose, audience, and ma ke
ocus. While he mas e –app en ice model s ill domina es music educa ion [10, 11],
echnology is inc easingly used bo h inside and ou side he class oom by adi ional
26 Chap e 2. Backg ound
Figu e 5: A Technology Mapping model showing he in e ela ions o musical ele-
men s and concep s, echnology, and a ec [156].
2.5. TEL Feedback Sys ems in Music Educa ion 27
and independen lea ne s. These ools ange om basic de ices like me onomes
and une s o ad anced sys ems ha p o ide pe sonalized eedback on pe o mance
[138]. This sec ion examines he cu en s a e o TEL o music pe o mance and
p ac ice o show how MuSA add esses a gap in his a ea.
2.5.1 Online Resou ces
In e ne A ailabili y. The a ailabili y o TEL in music educa ion would no be
possible wi hou he in e ne , which has been a powe ul ool in he dissemina ion o
in o ma ion, pa icula ly in egions whe e digi al access is widesp ead. In e ne con-
nec i i y has expanded ac oss households in he Uni ed S a es and Eu ope, making
hese egions pa icula ly ele an o examining echnology’s ole in music lea ning.
A 2023 Pew Resea ch Cen e su ey epo s ha 95% o U.S. adul s use he in e ne ,
90% own a sma phone, and 80% ha e high-speed in e ne a home [165]. Simila ly,
he Eu opean Commission epo ed in 2021 ha 96% o Eu opeans ha e access o
a mobile phone, and 82% o households ha e in e ne access [166]. These s a is ics
highligh he in as uc u al ounda ion o digi al lea ning ools in hese egions.
YouTube. A 2022 Pew Resea ch Cen e su ey ound ha YouTube4was he
mos widely used online ideo pla o m among Ame ican eenage s (96%) and Eu-
opean young adul s (84%). YouTube se es as a p ima y hub o in o mal beginne
lea ning ia ins uc ional mock ideo lesson demons a ing echnique, exe cises, and
epe oi e [138]. Ad anced lea ne s also use YouTube o access in o ma ion add ess-
ing echnical, philosophical, and ca ee challenges. While he pla o m’s s eng hs
lie in i s accessibili y, low cos , and sel -pacing h ough ideo con ols [138], i s majo
limi a ion is he lack o pe sonalized pe o mance eedback.
Music Educa ion Websi es. Music o ums and music educa ion websi es like
Da e Conse a oi e5, Music Theo y6, My Music Theo y7, and Sma Music8may ill
4h ps://you ube.com/
5h ps://da econse a oi e.o g/
6h ps:/music heo y.ne /
7h ps://mymusic heo y.com/
8h ps://sma music.com/
28 Chap e 2. Backg ound
YouTube’s gap by gene a ing con e sa ion abou echnique and heo y, bu wi hou
pe sonalized pe o mance eedback, hese pla o ms don’ add ess c i ical aspec s o
music alen de elopmen like pos u e, sound quali y, o musical exp ession [138].
Young adul lea ne s also a ely in e ac wi h hese pla o ms and seem o g a i a e
owa ds unidi ec ional con en [167], indica ing ha he con en le ied may be oo
edious o lea n wi hou a dedica ed ins uc o o a pla o m ha engages be e
wi h he mo i a ion o sel - egula ion p inciples o SDL.
Online Lea ning. Pe sonalized pe o mance eedback can be deli e ed emo ely
ia online lea ning, which includes synch onous, asynch onous, and hyb id o ma s.
Online lea ning is sugges ed o be as e ec i e as in-pe son ins uc ion [168, 169]
and may expand access o mas e -app en ice lea ning in domains like piano [170].
C i ics o online lea ning men ion ha eliance on SRL and s uden mo i a ion may
impac s uden engagemen [171, 172]. Rega dless, his emains a key challenge
o independen lea ning. Highe educa ion ins i u ions ha e adop ed ools ha
eplica e eal-wo ld musical wo kspaces such as LoLa9, JackT ip10, and Ul aG id11
o educe loca ion-based lea ning es ic ions and p omo e s uden independence.
Online Collabo a ion. Online collabo a ion is ano he way o gi e pe sonal-
ized pe o mance eedback and cul i a e mo i a ion. Pla o ms like AudioMo e s12,
Sou ce-Connec 13, Sessionwi e14, SonoBus15, and VST Connec 16 o e p oduc s wi h
p o essional-g ade audio s eaming o emo e collabo a ion [173]. These ools p i-
o i ize high-quali y sound, low-la ency capabili ies, and digi al audio wo ks a ion
in eg a ion, enabling emo e music c ea ion ha may ain c ea i e and exp essi e
skills, a key aspec o TPCK o e ec i e echnology implemen a ion in music ed-
uca ion [163]. Real- ime "jamming" ools like MusicianLink17 and A smesh18 can
9h ps://lola.con s.i /
10h ps://jack ip.com/
11h ps://ul ag id.cz/
12h ps://audiomo e s.com/
13h ps://sou ce-elemen s.com/p oduc s/sou ce-connec /
14h ps://sessionwi e.com/
15h ps://sonobus.ne /
16h ps://s einbe g.ne / s -connec /
17h ps://musicianlink.com/
18h ps://a smesh.com/
2.5. TEL Feedback Sys ems in Music Educa ion 29
also acili a e collabo a ion wi hou eliance on in-pe son sessions. Howe e , hese
high- ideli y ools ace challenges like la ency, da a loss, and high ba ie - o-en y.
2.5.2 Non-Real-Time Reco ding Feedback
Audio and Video Reco ding. Le e aging audio and ideo eco dings is ano he
way o ecei e pe sonalized pe o mance eedback and is p o en o imp o e one’s
own playing [61]. Many eache s al eady encou age s uden s o make ideo eco d
pe o mances o e iew mo emen s o ges u es ha a e di icul o no ice in eal- ime
due o he high cogni i e load o simul aneously pe o ming and sel -e alua ing [138].
Sepa a ing pe o mance and e iew enables mo e objec i e assessmen as eco dings,
whe he audio o ideo, allow musicians o analyze speci ic musical elemen s, com-
pa e eco dings wi h hose o eache s o p o essionals, and s udy physical echnique
in de ail [138]. Re iewing eco dings is also a ool o SRL and SDL by acili a -
ing be e musical unde s anding and se ing p ac ice goals [174]. Howe e , despi e
he bene i s o e iewing audio and ideo eco dings, many lea ne s ail o egula ly
e iew hei eco dings due o he ime consuming na u e o he p ocess [154]. S ud-
ies show ha using audio and ideo eco dings o p o ide compu e -assis ed isual
eedback can help musicians imp o e dynamic con ol and awa eness by gi ing clea
eedback on hei exp essi e ange [175].
Compu e -Assis ed Audio Reco ding Feedback. Audio eco dings can be
abs ac ed h ough he isualiza ion o impo an musical aspec s like equency
spec um, undamen al equency (pi ch), in ensi y, and esonance a eas ( o man s)
using ee audio analysis so wa e like P aa 19, Sonic Visualise 20, Audaci y21, o
LARA22 [176]. These ools also o m he basis o pu pose-designed isual eedback
ools ha display di e en audio analyses a di e en sc een loca ions, some imes
aligned wi h e e ence audio o in e p e esul s. Fo pi ched audio, hese include
19h ps:// on.hum.u a.nl/p aa /
20h ps://sonic isualise .o g/
21h ps:/audaci y eam.o g/
22h ps://hslu.ch/en/luce ne-school-o -music/ o schung/pe omance/la a/
30 Chap e 2. Backg ound
ools like Sing and See23, VocaVis a24, and Ta ini25, while o pe cussi e audio,
hey may include audio analysis so wa e plugins like Bea Roo o Onse DS om
Sonic Analyse [176]. Howe e , many o hese isual eedback ools ha e complica ed
in e aces wi h a high ba ie o en y o he a e age lea ne , may equi e an expe
wi h a sound and music compu ing backg ound o in e p e esul s, o a e p ima ily
o ien ed o eal- ime a he han non- eal- ime eedback.
Compu e -Assis ed Video Reco ding Feedback. Video eco dings can also
be abs ac ed ia isual eedback ools. Olde sys ems elied on ma ke s placed o
ack mo ion o ma k clea con as be ween an objec and i s backg ound, like Eye-
sWeb26, which analyzes exp essi e pe o mance. Pu due Uni e si y’s AIM p ojec
is de eloping an expe imen al ool called he E alua o 27 o assis s ing musicians
wi h indi idual p ac ice by analyzing a musician’s sound and ideo, compa ing i o
a digi ized sco e o de ec de ia ions in ele an musical aspec s (in ona ion, hy hm,
dynamics), and e en o e pos u e co ec ion sugges ions. Howe e , non- eal- ime
ideo eco ding pe o mance analysis so wa e specialized o musical pe o mances
has no ye been deeply explo ed. Gene al open-sou ce ideo analysis ools like
Kino ea28 a e a ailable o s udying body mechanics and ges u es. On o m29 is an-
o he ool, hough specialized o spo s pe o mance, ha allows ins uc o s o gi e
a ge ed eedback on ideo eco dings using oice-o e s and on-sc een ma kup ools.
2.5.3 Real-Time Feedback
Compu e -Assis ed Real-Time Audio Feedback. While a ailable ools o
compu e -assis ed audio eco ding eedback a e ei he expe imen al ools o aimed
a hose wi h p oduc ion, enginee ing, o sound and music compu ing backg ounds,
many ools ha e been de eloped o p o ide lea ne s eal- ime eedback based on
pe o mance audio. Some o hese ools include hose al eady men ioned such as
23h ps://singandsee.com/
24h ps:// oce is a.com/en/
25h ps://cs.o ago.ac.nz/g aphics/Geo / a ini/index.h ml
26h ps://casapaganini.unige.i /eyesweb_bp
27h ps://gi hub.com/Pu due-A i icial-In elligence-in-Music/E alua o -code
28h ps://kino ea.o g/
29h ps://on o m.com/
2.5. TEL Feedback Sys ems in Music Educa ion 31
EyesWeb, Sing and See, VocaVis a, and Ta ini. Gene al eal- ime eedback o ea
aining has been explo ed wi h Ea Mas e 30. Real- ime audio eedback o s ing
ins umen pe o mance has been pa icula ly explo ed wi h expe imen al open-
sou ce pla o ms In onia31, which helps s ing playe s isualize in ona ion. Mo e
ecen ly he SkyNo e32 web applica ion was de eloped o ack a s ing playe ’s pi ch
in eal- ime and o e lay hei pi ch cu e wi h a sco e, allowing lea ne s o see i hei
pi ch and iming co ec ly align wi h gi en no a ion. Plec us [177, 178] is ano he
expe imen al applica ion de eloped o aining in ona ion in no ce s ing playe s.
O he comme cial pla o ms le e aging eal- ime audio eedback include Yousician33,
MakeMusic34, Ube cho d35, and Tones o36, and Riyaz37 which all compa e use ’s
eal- ime audio o a selec ed sco e. Co osia38 by KORG p o ides eal- ime eedback
on ea u es such as dynamics, pi ch s abili y, imb e, and a ack cla i y, hough
ecei ing simul aneous aspec s can make i di icul o independen ly assess each
one.
Compu e -Assis ed Real-Time Video Feedback. Real- ime ideo eedback is
well illus a ed by he iMaes o p ojec , which c ea ed an “augmen ed mi o ” ha
uses isualiza ion and soni ica ion o analyze a s ing playe ’s ges u es li e. TELMI
is ano he p ojec o s ing playe s, o e ing eal- ime isual eedback on pos u e,
ges u e, and bowing, along wi h sco e-aligned ea u es such as dynamics, pi ch, and
imb e [179]. The E alua o ins ead compa es s uden pe o mances o p e- eco ded
examples wi h co ec o m. Howe e , eal- ime ideo eedback sys ems emain less
de eloped han audio ones, la gely due o he echnical demands o mo ion cap u e
and he cu en limi s o comme cial AI mo ion- acking so wa e.
30h ps://ea mas e .com/
31h ps://in onia.com/index.sh ml
32h ps://appskyno e.com
33h ps://yousician.com/
34h ps://makemusic.com/
35h ps://ube cho d.com/
36h ps:// ones o.com/
37h ps:// iyazapp.com/
38h ps://ko g.com/us/p oduc s/so wa e/co osia/
38 Chap e 3. Designing and Implemen ing MuSA
conside able o e lap in key exp essi e ea u es—such as in ona ion, ib a o, empo,
and dynamics—and also align wi h he p ima y esea ch a eas o my supe iso
( iolin) and co-supe iso ( oice).
3.2.1 Selec ing Musical Fea u es o MuSA-V2
One o he key pieces o eedback I ecei ed om he iolinis s a he RCM based
on MuSA-V1 was ha he mos impo an musical aspec s o isualize and indica e
momen s o saliency we e dynamics and empo. Bo h o hese aspec s ha e been
suppo ed in he li e a u e as indica o s o saliency h ough a iabili y [21, 25, 26].
Consequen ly, one o he main goals o he second e sion was o implemen a
eliable dynamics and empo ex ac ion algo i hm ha could also iden i y salien
momen s. This was a challenge in he o iginal e sion: al hough an onse en elope
was calcula ed using lib osa.onse .onse _s eng h, smoo hed wi h a median
il e , and a empog am was gene a ed wi h lib osa. ea u e. empog am, es ima -
ing and isualizing global and local empo using lib osa. ea u e. hy hm. empo
p o ed un eliable, as he empo cu e could no be accu a ely in e ed om he
empog am. The e o e, I needed o de elop a new app oach o isualizing empo
o e ime and selec ea u e ex ac ion algo i hms ha we e obus . Addi ionally,
I had o de e mine a se o musical aspec s o MuSA-V2 ha we e in e es ing o
implemen o bo h iolin and oice, while also na owing he scope o allow o a
comple e ede elopmen o he MuSA applica ion.
P io i izing some musical ea u es o e o he s. No all musical ea u es om
MuSA-V1 we e ca ied in o V2. While dynamics and pi ch emained, imb e and
a icula ion we e omi ed. In eg a ing hese was possible, bu my p io i y was de-
eloping a eliable dynamic and empo isualiza ion, and he RCM speci ically em-
phasized dynamics and empo.
Exclusion o imb e. Timb e is gene ally pa icula ly challenging o ex ac , as
i depends on in e ac ions among mul iple lowe -le el ea u es, including he ene gy
spec um, sho - e m ansien s, and co ela ions wi h loudness and pi ch [182, 183],
3.2. Implemen ing MuSA-V2 39
Table 2: Fea u e ex ac ion me hods and c i e ia o salien momen selec ion
(MuSA V2) wi h exac pa ame e s.
Musical Aspec Ex ac ion Me hod Saliency Selec ion
Dynamics F amewise RMS om lib osa.
ea u e. ms and smoo hed RMS
ace. Pa ame e s: N_FFT =
2048, HOP_LENGTH = 512,
smoo hing: mean il e wi h
window_pe cen age = 0.1. Sliding
window:
WINDOW_PERCENTAGE =
0.05, HOP_PERCENTAGE =
0.0125.
High- a iabili y sec ions compu ed
using s anda d de ia ion o e
windows. Th eshold = 50% o
max a iabili y. Re u ns ame
indices and audio- ime anges.
Pi ch F amewise pi ch in Hz wi h
con idence, smoo hed and il e ed
by RMS. Ex ac ed using CREPE
(Essen iaPi chCREPEmodel,
CREPE_HOP_DURATION_SEC
= 0.01) o Lib osa pyin.
Smoo hing/segmen a ion:
WINDOW_PERCENTAGE =
0.05, HOP_PERCENTAGE =
0.0125, RMS h eshold = 0.01.
High- a iabili y sec ions compu ed
using
calcula e_high_ a iabili y_sec ions
wi h is_pi ch=T ue. Sec ions
wi h la ges de ia ion om
nea es equal- empe ed piano no e
we e highligh ed.
Tempo Dynamic empo in BPM using
Essen ia
Bea T acke Mul iFea u e.
Smoo hed and in e pola ed o
egula ime axis. Sample a e =
44100 Hz. Smoo hing: median
il e . Window ac ion: 0.1.
No disc e e highligh ed sec ions;
e u ns con inuous in e pola ed
empo ace o e ime.
Phona ion F ame/window-le el phona ion
class p obabili ies (b ea hy, low,
neu al, p essed) using VGGISH
embeddings and
PHONATION_MODEL (Ke as).
Window size = 2 ames, hop = 1
ame, ame du a ion = 0.1 s.
No highligh ed sec ions; e u ns
pe -class p obabili y ime se ies.
Vib a o Pe - ame ib a o oscilla ions
a ound smoo hed pi ch. De ec ion
uses s able-no e iden i ica ion and
oscilla ion checks. Vib a o ex en
(ampli ude) and a e (Hz)
calcula ed pe sec ion. Th esholds:
min_sus ained_leng h = 5
ames, gap h eshold = 5 ames.
Ex en h eshold o highligh ing
= 1.9. CREPE_HOP_
DURATION_SEC = 0.01 used o
ime con e sion.
Highligh ed sec ions a e ames
wi h ex en > 1.9, g ouped in o
consecu i e segmen s. Re u ns
bo h ame anges and audio- ime
anges.
40 Chap e 3. Designing and Implemen ing MuSA
and so I chose o exclude i om MuSA-V2. A icula ion was also le aside due o
ime cons ain s. Based on RCM eedback, empo was added, along wi h ib a o
and phona ion mode o oice, guided by he wo k o my co-supe iso Su i Hää ä,
wi h ib a o included because i is ele an o bo h oice and iolin. The inal se o
musical aspec s, hei ex ac ed audio ea u es, and he me hod o selec ing salien
momen s a e summa ized in Table 2.
Implemen ing dynamic window and hop sizes. An impo an goal was o
ensu e ha ex ac ed ea u e cu es main ained consis en smoo hing and da a ep-
esen a ion ega dless o audio leng h. Fo ins ance, a sho e audio segmen using a
ixed window size would yield ewe da a poin s han a longe segmen , which could
lead o noisie cu es i he window size we e no adjus ed. To add ess his, sliding
window and hop sizes we e de ined as pe cen ages o he audio leng h whe e applica-
ble (e.g., o dynamics and pi ch), allowing he ea u e cu es o scale p opo ionally
and emain compa able ac oss eco dings.
S uggles wi h iden i ying a iabili y h esholds. One key piece o eedback
om MuSA-V1 was he need o iden i y mul iple salien momen s wi hin each audio
ea u e, a he han a single ins ance. This was add essed by selec ing all momen s
exceeding a speci ic a iabili y h eshold. Al e na i e app oaches could ha e in-
cluded selec ing he op nmos a iable sec ions o applying a h eshold o he op
nsec ions, bu he implemen ed me hod was chosen o a oid excluding po en ially
in e es ing momen s. The h esholds we e de e mined empi ically du ing de el-
opmen a he han being d awn om exis ing li e a u e, as ew sou ces p o ided
guidance on p ecise alues. Es ablishing h esholds based on p io s udies ep esen s
a clea a enue o imp o ing he obus ness and gene alizabili y o he me hod in
u u e wo k.
S uggles wi h empo ex ac ion. Dynamic empo acking was a majo chal-
lenge o his p ojec . I es ed se e al algo i hms, including Essen ia’s Rhy hmEx ac o 2013
and Tenso lowP edic TempoCNN. Because oice and iolin lack a s ong bea ,
Rhy hmEx ac o 2013 e u ned ew usable poin s, while Tenso lowP edic TempoCNN
3.2. Implemen ing MuSA-V2 41
p oduced only 2–3 empo alues pe segmen , especially o sho audio, limi ing
a iabili y analysis. Bea T acke Mul iFea u e p o ided mo e eliable es ima es
and a highe numbe o poin s, so i was selec ed o implemen a ion. Howe e ,
he densi y o poin s was s ill insu icien o con iden ly iden i y salien momen s,
so empo a iabili y was no ully in eg a ed in o MuSA-V2. In es iga ing op imal
empo ex ac ion me hods emains an in e es ing di ec ion o u u e esea ch, bu
i ell ou side he scope o his hesis, which ocused on designing and building he
inal MuSA-V2 web applica ion.
Implemen ing oice-speci ic ea u es. Phona ion mode was implemen ed as
an expe imen al ea u e in MuSA-V2. Phona ion mode desc ibes he manne in
which he ocal olds ib a e o p oduce sound, and changes in phona ion h ough-
ou a pe o mance can p o ide aluable insigh s o singe s, pa icula ly in e ining
exp essi e echniques and con eying emo ion. Su i Hää ä de eloped an expe i-
men al phona ion mode classi ica ion model using he Yesile Phona ion da ase ,
dis inguishing ou ocal modes: b ea hy (so , ai y), neu al (s anda d singing o
speech), low ( esonan and clea ), and p essed ( ense, ha sh). I inco po a ed i-
sualiza ions o hese phona ion modes in o MuSA wi h he in en ion o explo ing
salien momen s based on mode ansi ions; howe e , due o ime cons ain s, his
unc ionali y was no ully implemen ed. Ne e heless, all ou phona ion modes
can be isualized in he applica ion, ep esen ing a p omising di ec ion o u u e
esea ch.
Fea u e ex ac ion akeaways. O e all, ea u e ex ac ion was a ounda ional
componen in de eloping MuSA-V2, enabling se e al success ul salien momen i-
sualiza ions. Fo example, dynamics p oduced clea highligh ed sec ions, which may
be suppo ed by indings om he e alua ion s udy (Sec ion 4). Pi ch ex ac ion
using CREPE esul ed in accu a e, smoo h pi ch cu es, and he highligh ed sec ions
e ec i ely cap u ed momen s o po en ial in ona ion—whe e equencies de ia ed
om hei co esponding equal- empe ed piano no es. Vib a o ex ac ion was ela-
i ely s aigh o wa d, hough imp o emen s could be made o iden i y momen s o
a iabili y mo e speci ically, a he han elying solely on a ixed h eshold. These
42 Chap e 3. Designing and Implemen ing MuSA
and o he limi a ions o salien momen iden i ica ion, as well as po en ial al e na i e
app oaches, a e discussed in Sec ion 5.1.4.
3.2.2 Designing MuSA
In designing MuSA-V2, I conside ed lea ne -cen e ed eaching (LCT), sel - egula ed
lea ning (SRL), and sel -di ec ed lea ning (SDL) p inciples (Sec ion 2.3), alongside
insigh s om Technological Pedagogical Con en Knowledge (TPCK; Sec ion 2.4).
TPCK highligh ed he impo ance o iden i ying opics ha a e di icul o each
o unde s and wi hou echnology, and o p esen ing con en in ways only possible
wi h echnological ools. Acco dingly, MuSA-V2 was designed o suppo bo h in-
dependen and class oom lea ning, using a web applica ion o make salien momen
iden i ica ion and isual eedback on audio eco dings mo e in ui i e. No ably, algo-
i hmically de ec ing salien momen s based on a iabili y is only easible wi h such
echnological suppo .
MuSA landing page. Based on c i iques o MuSA-V1, I aimed o edesign he
in e ace o minimize he numbe o s eps equi ed, educing ex aneous cogni i e
load o use s. Fo example, use s i s selec one o h ee ins umen s on he land-
ing page— iolin, oice, o polyphonic (Figu e ??). The polyphonic op ion allows
use s o analyze eco dings wi h mul iple ins umen s o melodic lines, making he
applica ion mo e b oadly accessible. Once an ins umen is selec ed, he a ailable
analyses a e au oma ically il e ed o hose ele an o ha ins umen . Addi ion-
ally, audio can be eco ded di ec ly wi hin he web applica ion, elimina ing he
need o use sepa a e pla o ms, al hough use s can s ill upload p e- eco ded audio
i p e e ed (Figu e ??).
Fea u e-speci ic isualiza ions. Addi ionally, i was impo an o ensu e ha
he musical aspec isualiza ions we e p ope ly con ex ualized agains hei own
scales (in a musically ele an way, no jus compu a ionally), so use s would be able
o ge meaning ou o he esul s wi hou dealing wi h he cogni i e load associa ed
wi h ying o in e p e e y nume ical esul s. Fo example, pi ch (Figu e 8b)
3.3. MuSA Sys em A chi ec u e 43
and ib a o (Figu e 8d) a e displayed wi h a piano- oll backg ound o con ex ualize
equency wi hin musical no a ion. Dynamics (Figu e 8a) a e p esen ed wi h simple
ampli ude o BPM scales. The di e en phona ion mode classes (Figu e 8c) a e
shown in sepa a e windows o e lec i ’s dimensional na u e. Vib a o also has
sepa a e windows o ex en and a e.
Salien momen isualiza ions. Salien momen s, when a ailable o a speci ic
ea u e, a e highligh ed and synch onized wi h he audio wa e o m. Use s can click
di ec ly on a highligh ed sec ion o play he co esponding audio, p o iding an im-
media e connec ion be ween he audio and isual elemen s o he applica ion. Zoom
and pan unc ionali y is a ailable ac oss all g aphs, ensu ing ha highligh ed sec-
ions and audio playback emain aligned a any le el o de ail.
O he conside a ions. Use s who a e un amilia wi h MuSA-V2 equi e guidance
on how o na iga e he in e ace and iden i y he unc ions o each bu on. To
add ess his, in o ma ion ool ips a e oggleable a he han always displayed. Since
he applica ion in ol es sending audio eco dings o a se e o analysis, use s can
also choose whe he hei da a is collec ed ia a oggleable consen bu on (de aul
on). Addi ionally, ea u e calcula ions a e cached o a oid edundan compu a ions,
imp o ing he o e all use expe ience (discussed u he in he ollowing sec ion).
3.3 MuSA Sys em A chi ec u e
Ano he cen al conside a ion in building MuSA-V2 was ensu ing accessibili y and
scalabili y so ha he applica ion could meaning ully impac a wide ange o use s.
To suppo deploymen a scale and simul aneous mul i-use access, he sys em was
designed wi h an a chi ec u e comp ising ou main componen s. Figu e 10 p o ides
an o e iew o he sys em a chi ec u e.
•F on end (Reac 4+ Tailwind5): Handles he in e ac i e use in e ace and
ende s ea u e isualiza ions using D36. The componen -based model o Reac
4h ps:// eac .de /
5h ps:// ailwindcss.com/
6h ps://d3js.o g/
44 Chap e 3. Designing and Implemen ing MuSA
Figu e 7: MuSA landing page ( op) and ile selec ion o audio eco ding (bo om).
3.3. MuSA Sys em A chi ec u e 45
(a) Dynamics iew. (b) Pi ch iew.
(c) Phona ion iew. (d) Vib a o iew.
(e) Tempo iew.
Figu e 8: O e iew o MuSA analysis iews.
46 Chap e 3. Designing and Implemen ing MuSA
and he u ili y- i s s yling o Tailwind acili a e a lexible and main ainable
design.
•Node.js7API: Se es as he middle laye be ween he on end and he Mon-
goDB8da abase, managing session me ada a and communica ion.
•Py hon Se ice (Flask9+ Gunico n10): Manages all audio- ela ed asks,
including p ep ocessing, signal and ea u e ex ac ion, saliency analysis, and
caching. Redis11 is used o sho - e m caching o minimize edundan com-
pu a ion, and audio iles a e s o ed locally on he se e .
•In as uc u e (Ubun u + Nginx12): Handles da a managemen , deploy-
men , and use au hen ica ion. Ubun u is he se e wi h Nginx con igu ed
as a e e se p oxy. Redis manages caching, MongoDB s o es use and session
me ada a, and JSON Web Tokens (JWT)13 secu e use sessions.
Explana ion o a chi ec u e. A sepa a e Py hon se ice was used ins ead o
Essen ia.js o se e al easons. Essen ia.js is op imized o sho clips (unde one
minu e), while his applica ion a ge s ull pe o mance eco dings. I s unc ionali y
is also mo e limi ed; o example, Essen ia’s CREPE pi ch- acking implemen a ion
is no a ailable in Essen ia.js. Finally, se e -side p ep ocessing—such as imming
silences wi h Lib osa and esampling o he equi ed sample a e—was necessa y
be o e analysis. Admi edly, a clien -only app oach using Essen ia.js migh ha e
su iced o a p o o ype on sho audio segmen s, bu his was a lesson lea ned in
hindsigh . To ensu e use p i acy, da a is main ained in isola ion h ough he use o
Redis-based sho - e m caching and JWT au hen ica ion, ensu ing ha each use ’s
uploaded audio and co esponding analyses emain p i a e.
Example use in e ac ion. To be e unde s and how he a chiec u e wo ks, a
p ocess diag am o an example use low is shown in Figu e 9. When a musician
7h ps://nodejs.o g/
8h ps://mongodb.com/
9h ps:// lask.palle sp ojec s.com/en/s able/
10h ps://gunico n.o g/
11h ps:// edis.io/
12h ps://nginx.o g/
13h ps://jw .io/in oduc ion#wha -is-json-web- oken
3.3. MuSA Sys em A chi ec u e 47
i s logs in o MuSA, hei session is au hen ica ed h ough a secu e JWT connec-
ion managed by a Redis cache. Once inside, hey selec he ins umen hey wan
o analyze— o example, iolin. The sys em hen p omp s hem o choose which
pe o mance ea u es o ocus on, such as pi ch s abili y, loudness, o a icula ion.
A e uploading hei eco ding, he Py hon se ice p ocesses he audio, ex ac s
he chosen ea u es, calcula es saliency, and caches he esul s o speed up in e ac-
ions. The use hen explo es hese esul s in he Reac on end, whe e hey can
play back hei pe o mance, zoom in o speci ic passages, and see salien momen s
highligh ed di ec ly on he imeline. When hey a e inished, hei analysis is s o ed
au oma ically, wi h me ada a sa ed in MongoDB and he audio p ese ed on he
se e so hey can e u n o i in u u e p ac ice sessions.
1. Au hen ica ion: JWT session wi h Redis cache
2. Ins umen Selec ion:
Violin, Voice, o Polyphonic
3. Fea u e Selec ion: Choose ea u es o analyze
4. Fea u e Ex ac ion: Py hon se ice an-
alyzes audio, calcula es saliency, caches esul s
5. Visualiza ion: Reac on end
ende s da a, synch onizes playback,
zoom, and highligh ed salien momen s
6. S o age: Me ada a in MongoDB, audio on se e
Figu e 9: Example use in e ac ion wi h MuSA.
54 Chap e 4. MuSA E alua ion S udy
Figu e 12: Dis ibu ion o he 14 pa icipan s e ained o inal analysis ac oss mu-
sical expe ience le els: 1 Beginne , 7 In e media e, 5 Ad anced, and 1 P o essional.
When mapped on o he TAD Music Model (Ap i ude, Compe ence, Expe ise,
T ans o ma ional Achie emen ), he sample e lec s a concen a ion in he Com-
pe ence (In e media e) and Expe ise (Ad anced) s ages, aligning wi h he sound
and music compu ing s uden demog aphic a Uni e si a Pompeu Fab a.
a ion design allows o a mo e di ec compa ison be ween pe o mance ou comes
and unde lying psychological and cogni i e p ocesses. Al hough amilia labels such
as In e media e and Ad anced we e used o desc ibe pa icipan s, hese ca ego ies
co espond o he Compe ence and Expe ise s ages o he TAD amewo k. This
mapping makes i possible o in e p e di e ences in sel -pe cep ion and pe o mance
no only in e ms o musical skill bu also h ough he lens o de elopmen al p o-
cesses in expe ise acquisi ion. Fo ins ance, pa icipan s a he Compe ence s age
may ely mo e hea ily on ex e nal eedback o s uc u e p ac ice and e ine hei
playing, whe eas hose a he Expe ise s age may in eg a e eedback di e en ly,
d awing on in e nalized s a egies and p io knowledge.
4.2.1 Objec i e Pe o mance Me ics
Objec i e pe o mance was de i ed om compu a ional analysis o eco ded audio
using he same ea u e ex ac ion algo i hms implemen ed in MuSA. Pi ch was es i-
4.2. Resul s 55
ma ed using CREPE [189], dynamics we e quan i ied ia Roo Mean Squa e (RMS)
ene gy using he ea u e. ms unc ion om lib osa, and empo was measu ed
using Essen ia’s Bea T acke Mul iFea u e algo i hm [190, 191].
Pa icipan eco dings unde con ol and eedback condi ions we e compa ed wi h
e e ence pe o mances. To accoun o empo al a iabili y, dynamic ime wa p-
ing (DTW) aligned pe o mances o poin -by-poin compa ison, a echnique ha
has been p o en o wo k e ec i ely o aligning audio signals unde a a ie y o
condi ions [192, 193]. Because pi ch, dynamics, and empo ope a e on di e en
scales, analyses we e conduc ed in h ee complemen a y ways: aw RMSE alues o
unp ocessed de ia ions, baseline-s anda dized pe cen age imp o emen s o wi hin-
subjec changes, and z-sco e s anda diza ion o allow c oss- ea u e compa ison and
e ec size e alua ion. E ec sizes we e calcula ed using Cohen’s d, which quan i ies
he s anda dized mean di e ence be ween g oups, p o iding a measu e o p ac ical
signi icance beyond s a is ical es s. This app oach p o ides bo h ea u e-speci ic
and in eg a i e pe spec i es on he impac o MuSA eedback.
Raw RMSE esul s. As shown in Figu e 13, aw RMSE alues be o e and a -
e he in e en ion displayed conside able sp ead ac oss all h ee musical ea u es.
Tempo was he only ea u e wi h a s a is ically signi ican e ec , wi h he eedback
g oup pe o ming wo se han he con ol g oup (p= 0.0096,d= 0.97; Appendix
Table 4). Pi ch and dynamics showed no signi ican condi ion e ec s (p > 0.32,
|d|<0.27), hough he wide a iabili y illus a ed in he boxplo s sugges s subs an-
ial indi idual di e ences ha may ha e masked smalle e ec s. Taken oge he ,
hese esul s sugges ha empo was mos sensi i e o dis up ion om eedback,
while pi ch and dynamics emained inconclusi e a his le el o analysis (see Ap-
pendix Table 5).
Expe ience-le el analyses u he cla i ied hese aw RMSE pa e ns. As shown in
Figu e 14, in e media e musicians showed mode a e e ec s o pi ch (∆ = −77.05,
d=−0.37), small e ec s o dynamics (∆=0.0038,d= 0.15), and mode a e e -
ec s o empo (∆ = 5.71,d= 0.38), all co esponding o nonsigni ican esul s
56 Chap e 4. MuSA E alua ion S udy
Figu e 13: Raw RMSE pe o mance ac oss h ee musical ea u es—pi ch (Hz), dy-
namics (RMS), and empo (BPM)—be o e and a e he in e en ion o bo h con-
ol and eedback g oups. Box plo s display medians, qua iles, and ou lie s o each
condi ion. Lowe RMSE alues indica e be e pe o mance. Sample sizes anged
om 9–14 pa icipan s pe condi ion. Resul s show subs an ial sp ead ac oss pa -
icipan s, wi h empo mos clea ly dis up ed by eedback, while pi ch and dynamics
show no clea condi ion-le el e ec s.
Figu e 14: Change in RMSE o pi ch, dynamics, and empo ac oss ou musician
expe ience le els (A e – Be o e), wi h nega i e alues indica ing imp o ed pe o -
mance. No s a is ically signi ican e ec s we e obse ed o condi ion o iming (all
p > 0.05). Beginne and P o essional g oups had e y small sample sizes (n=2),
and al hough some la ge e ec sizes we e seen (e.g., In e media e empo d= 0.94,
Beginne dynamics d= 7.58), hese esul s a e likely in luenced by high a iabili y.
(p > 0.09). Ad anced musicians exhibi ed small- o-mode a e e ec s ac oss ea u es
(pi ch: ∆ = 14.65,d= 0.33; dynamics: ∆ = −0.0048,d=−0.16; empo: ∆=9.54,
d= 0.59), again nonsigni ican (p > 0.32). Beginne and p o essional g oups had ex-
emely small sample sizes, p oducing highly a iable and uns able e ec sizes (e.g.,
Beginne dynamics Feedback s Con ol: d= 7.58) ha we e also nonsigni ican .
O e all, al hough no compa isons eached s a is ical signi icance, he e ec sizes
and mean di e ences sugges ha esponses o eedback a ied by expe ience, wi h
4.2. Resul s 57
in e media e musicians ending owa d mode a e imp o emen s and beginne s/p o-
essionals showing high a iabili y and ex eme bu un eliable e ec s (Appendix
Tables 6-8).
Baseline-s anda dized imp o emen s. To accoun o baseline pe o mance
di e ences, pe cen imp o emen sco es we e calcula ed o each ea u e (Appendix
Table 9, Figu e 15). Ac oss all pa icipan s, eedback showed a modes ela i e ad-
an age o dynamics (+14.1 pp compa ed o con ol), bu no bene i o pi ch (–10.1
pp) and a ma ked de imen o empo (–110.3 pp). None o hese con as s eached
s a is ical signi icance (p > 0.28), e lec ing bo h limi ed sample size and high in e -
indi idual a iabili y, bu he pa e n sugges s ha eedback selec i ely in luenced
empo al con ol while o e ing limi ed suppo o pi ch o dynamic accu acy.
Figu e 15: Pe cen imp o emen om indi idual baselines ac oss pi ch, dynamics,
and empo. Violin plo s wi h o e laid box plo s show pe cen age change ela i e
o p e-in e en ion baseline. Ho izon al dashed line indica es no change. S a is ical
ma ke s show e ec sizes and p- alues (all non-signi ican ).
When analyzed by musical expe ience le el (Appendix Table 10, Figu e 16), no
s a is ically signi ican subg oup e ec s we e obse ed. Desc ip i ely, in e media e
musicians showed he la ges posi i e di e ences ac oss ea u es (dynamics +27.1%,
empo +116.7%, pi ch +2.9%), while ad anced musicians’ ou comes ended close
o ze o o nega i e (pi ch –28.5%, empo –392.0%, dynamics +0.3%). Beginne s and
p o essionals con ibu ed e y limi ed da a, wi h single cases pe condi ion, making
hei esul s unin e p e able.
58 Chap e 4. MuSA E alua ion S udy
Figu e 16: Pe cen imp o emen om indi idual baselines by musical expe ience
le el (Beginne , In e media e, Ad anced, P o essional). Violin plo s compa e eed-
back and con ol g oups wi hin each ca ego y; all compa isons we e non-signi ican .
Dashed line shows no change om baseline.
Z-sco e s anda dized compa isons. To accoun o c oss- ea u e scale di e -
ences, RMSE alues we e z-sco e s anda dized agains baseline dis ibu ions. As
shown in Figu e 17, eedback was associa ed wi h a small posi i e shi in dynamics
(+0.41 SD, d= 0.41) and a negligible e ec o pi ch (–0.17 SD, d= 0.16), while
empo showed a small de imen al shi (–0.43 SD, d= 0.43). None o hese di e -
ences eached s a is ical signi icance (p > 0.28). Mo e summa y s a is ics, including
e ec sizes and p- alues, a e p o ided in Appendix Tables 11–12.
Figu e 17: Z-sco e s anda dized imp o emen by ea u e (Pi ch, Dynamics, Tempo)
o con ol and eedback condi ions. Violin plo s show dis ibu ions, wi h o e laid
box plo s o median and qua iles. Posi i e alues indica e abo e-baseline imp o e-
men . All compa isons we e non-signi ican (p > 0.28).
When b oken down by musical expe ience (Figu e 18), no subg oup e ec s eached
s a is ical signi icance, hough desc ip i e di e ences appea ed. In e media e musi-
4.2. Resul s 59
Figu e 18: Z-sco e s anda dized imp o emen by expe ience le el (Beginne , In e -
media e, Ad anced, P o essional) o Pi ch, Dynamics, and Tempo. Violin plo s
show dis ibu ions o con ol and eedback condi ions. S a is ical anno a ions indi-
ca e e ec sizes (d) and p- alues. Mos di e ences we e non-signi ican , highligh ing
he e ogenei y in esponses ac oss expe ience le els.
cians showed he la ges posi i e shi s (dynamics +0.78 SD, empo +0.46 SD, pi ch
+0.05 SD), while ad anced musicians ended owa d neu al o nega i e changes
(pi ch –0.46 SD, empo –1.54 SD, dynamics +0.01 SD). Beginne and p o essional
pa icipan s con ibu ed only isola ed da a poin s, limi ing in e p e abili y. O e all,
hese s anda dized analyses sugges po en ial a ia ion in eedback e ec s by expe-
ience le el, bu gi en small samples and non-signi ican es s, hese ends should
be conside ed explo a o y.
4.2.2 Pe cei ed Pe o mance Me ics
To e alua e pa icipan s’ subjec i e expe ience o imp o emen , sel - a ings on pi ch,
dynamics, and empo we e collec ed be o e and a e p ac ice sessions unde con ol
(no eedback) and eedback condi ions. Pai ed - es s we e conduc ed wi hin each
condi ion o assess changes o e ime, while be ween-condi ion compa isons ( eed-
back s. con ol) we e pe o med on p e- and pos -p ac ice a ings. E ec sizes
(Cohen’s d) we e calcula ed o quan i y he magni ude o obse ed di e ences. This
app oach allowed o examining bo h o e all pe cei ed pe o mance changes and
a ia ions ac oss musical expe ience le els, p o iding a complemen a y pe spec i e
o objec i e pe o mance me ics.
O e all, pa icipan s’ sel - a ings a ied by ea u e and condi ion (Figu e 19). Fo
60 Chap e 4. MuSA E alua ion S udy
Figu e 19: Compa ison o sel - a ed pe o mance be o e and a e he in e en ion
ac oss di e en musical expe ience le els. Each subplo ep esen s one musical ea-
u e (pi ch, dynamics, empo). Box colo s dis inguish ’Be o e’ and ’A e ’ a ings,
highligh ing changes in pe cei ed pe o mance ac oss expe ience le els.
pi ch, pe cei ed imp o emen s we e minimal, wi h eedback p oducing a small a -
e age inc ease (+0.15 poin s, p= 0.656,d=−0.122), indica ing li le impac o
he eedback ool. Dynamics showed he la ges pe cei ed gains, inc easing on a -
e age by +1.00 poin s wi h eedback, which was highly signi ican (p= 0.0009∗∗∗,
d=−1.003). Tempo imp o emen s we e less consis en : eedback led o a mode a e
inc ease (+0.58 poin s, p= 0.089,d=−0.563), while p ac ice alone p oduced a
sligh ly la ge imp o emen (+0.94 poin s, p= 0.0106∗), sugges ing gene al p ac ice
con ibu ed o empo pe cep ion mo e han eedback (Appendix Table 13).
Conside ing musical expe ience (Figu e 21), di e ences eme ged ac oss ea u es.
Fo pi ch, ad anced pa icipan s showed he la ges pe cei ed imp o emen wi h
eedback (+0.60), hough his change was no s a is ically signi ican (p= 0.4766,
d= 0.14), whe eas beginne s showed inconsis en changes. Dynamics bene i ed
all expe ience le els, wi h beginne s and in e media es showing he la ges posi-
i e shi s (+2.00 and +1.29 poin s wi h eedback, espec i ely); only in e media es
eached s a is ical signi icance (p= 0.0004∗∗∗,d= 0.92), while changes o beginne s
4.2. Resul s 61
Figu e 20: Sel - epo ed pe o mance a ings by musical expe ience ac oss h ee
ea u es: (A) Pi ch, (B) Dynamics, (C) Tempo. Box plo s show a ings be o e
(ligh pu ple) and a e (medium pu ple) p ac ice, combining con ol (no eedback)
and expe imen al ( eedback) condi ions. Medians ( hick line), in e qua ile ange
(box), and ou lie s (ci cles) a e shown. Expe ience le els: Beg = Beginne , In =
In e media e, Ad = Ad anced, P o = P o essional.
Figu e 21: Mean pe cei ed imp o emen (a e minus be o e) by musical expe ience
and in e en ion ype o (A) Pi ch, (B) Dynamics, (C) Tempo. Da k pu ple ba s
indica e eedback, ligh pu ple ba s indica e con ol. Posi i e alues e lec imp o e-
men , nega i e alues e lec decline. Numbe s abo e ba s show exac alues ≥0.01.
Ho izon al line a ze o indica es no change.
(+2.00) and ad anced pa icipan s (+0.40) we e no signi ican . Tempo imp o e-
men s we e mos no able in he con ol condi ion o beginne s (+3.00, p= 0.011∗),
wi h smalle gains om eedback obse ed in o he g oups (in e media e: +1.05,
p= 0.089; ad anced: +0.00, p= 1.0). O e all, pe cep ion o dynamics pe o mance
imp o ed wi h MuSA, pa icula ly o in e media e pa icipan s, while pe cei ed
imp o emen s in pi ch we e limi ed and empo gains likely e lec ed gene al p ac ice
e ec s a he han MuSA’s e lec i e eedback (Appendix Table 14).
62 Chap e 4. MuSA E alua ion S udy
4.2.3 Sel -Awa eness Me ics
Sel -awa eness e lec s he deg ee o which pa icipan s’ pe cep ions o hei own
pe o mance align wi h objec i ely measu ed ou comes. To quan i y his, pe cei ed
imp o emen s we e compa ed wi h ac ual pe o mance changes using Pea son co -
ela ion coe icien s. Co ela ions we e calcula ed be ween sel - epo ed a ings and
RMSE alues, be ween pe cei ed imp o emen and pe cen age imp o emen , and
be ween help ulness a ings and ac ual pe o mance gains, p o iding a p ima y in-
dex o sel -assessmen accu acy.
Fo di ec compa ison ac oss scales, objec i e imp o emen s we e escaled o ap-
p oxima ely ma ch he Like ange by di iding by 7. Absolu e sel -awa eness e o
was hen compu ed as he absolu e di e ence be ween subjec i e and scaled objec-
i e imp o emen s, wi h lowe alues indica ing g ea e accu acy. Condi ion-le el
di e ences we e assessed using Mann–Whi ney U es s, independen - es s, and
Cohen’s d o quan i y e ec sizes.
O e all me ics. Compa ing subjec i e sel - a ings o objec i e measu es ac oss
pi ch, dynamics, and empo e ealed gene ally weak co ela ions, indica ing poo
sel -awa eness among pa icipan s. O e all, he Con ol g oup showed =−0.027
(p= 0.874,n= 36) and he Feedback g oup =−0.182 (p= 0.304,n= 34),
wi h nei he eaching s a is ical signi icance. Fea u e-speci ic analyses mi o ed his
pa e n: o pi ch, = 0.089 (Con ol) and =−0.160 (Feedback), p= 0.062;
o dynamics, =−0.267 (Con ol) and = 0.101 (Feedback), p= 0.148; and o
empo, = 0.317 (Con ol) and =−0.363 (Feedback), p= 0.694. None o hese
co ela ions we e s a is ically signi ican , sugges ing ha pa icipan s’ pe cei ed im-
p o emen s did no eliably align wi h objec i e pe o mance ac oss ea u es o con-
di ions (Appendix Table 15).
To u he cha ac e ize sel -awa eness, pa icipan s’ assessmen s we e classi ied in o
ou ca ego ies: Accu a e,O e con iden ,Unde con iden , and Somewha
Inaccu a e. Classi ica ion was based on he di e ence be ween subjec i e imp o e-
men and scaled objec i e imp o emen . A esponse was conside ed Accu a e i his
4.2. Resul s 63
Figu e 22: Co ela ion sco es be ween sel - a ings and objec i e pe o mances o
h ee musical ea u es (pi ch, dynamics, and empo).
Figu e 23: Sca e plo s examining he ela ionship be ween pa icipan s’ pe cei ed
imp o emen (subjec i e sel - a ings) and ac ual pe o mance imp o emen (objec-
i e measu emen s) ac oss h ee musical ea u es: pi ch, dynamics, and empo. Con-
ol pa icipan s a e shown in o ange, and eedback pa icipan s in blue. The ed
dashed line indica es pe ec sel -awa eness.
absolu e di e ence was ≤1, accoun ing o ypical pe cep ual noise. O e con iden
esponses e lec ed la ge posi i e subjec i e changes despi e pe o mance decline
(subjec i e imp o emen >1and objec i e imp o emen <−5%), whe eas Unde -
con iden esponses e lec ed nega i e sel - a ings despi e subs an ial imp o emen
(subjec i e imp o emen <−1and objec i e imp o emen >10%). Remaining
cases we e labeled Somewha Inaccu a e, cap u ing mode a e misalignmen s wi h-
70 Chap e 5. Discussion
decisions. By highligh ing salien pe o mance sec ions and showing ea u e-le el de-
ia ions in pi ch, dynamics, o iming, MuSA enables lea ne s o a ge hese a eas
o pa icula ou comes based on hei in ended exp ession. In his way, he ool sup-
po s sel -di ec ed lea ning (SDL) and lea ne -cen e ed eaching (LCT) p inciples,
gi ing s uden s owne ship o e how hey analyze and e ine hei pe o mances, and
allowing exp essi e goals o eme ge as a p oduc o in o med, ea u e-d i en p ac ice
a he han p esc ip i e ins uc ion.
The concep o salience analysis was cen al o MuSA’s design. Ra he han p e-
sc ibing how a musician should exp ess hemsel es, MuSA iden i ies momen s in a
pe o mance ha de ia e om ea u e-speci ic expec ed no ms. These no ms a e
compu ed ela i e o he uploaded pe o mance i sel . Fo example, a as dynamic
change o high a iabili y in dynamics compa ed o su ounding sec ions would be
highligh ed. Fo pi ch, salien momen s co espond o de ia ions om he expec ed
no e wi hin a 12-semi one scale, and ib a o sec ions a e de ec ed and highligh ed
when hey exceed a de ined h eshold. Impo an ly, hese expec ed no ms a e in-
he en ly lexible and can be adap ed based on a use ’s goals, cul u al con ex , o
no a ion p e e ences.
As a poin o compa ison, one o he goals o TELMI, he pa en p ojec o MuSA,
was o p o ide eal- ime isual eedback on imb e, a ea u e ha is inhe en ly
subjec i e. The success o TELMI demons a es ha meaning ul eedback can be
p o ided e en when he unde lying musical a ibu e is di icul o quan i y, and
ha i is bo h accep able and expec ed o TEL ools o in e p e measu emen s o
classi ica ions o subjec i e da a in o de o o e a ge ed eedback o lea ne s [1].
Simila ly, by de ining salien momen s h ough a speci ic, ea u e-d i en app oach,
MuSA helps na ow he lea ne ’s ocus du ing p ac ice. Ins ead o ha ing o conside
all possible ways a momen could be in e p e ed as salien , he ool p o ides one
conc e e lens o a en ion, guiding p ac ice in a manageable and s uc u ed way.
This does no p eclude he use o o he echniques, such as human-guided eedback
based explici ly on a ec , which could be applied alongside MuSA o cap u e aspec s
o pe o mance ha ex end beyond ea u e-based salien momen analysis.
5.1. Discussion 71
5.1.2 E alua ing MuSA: Demand, Limi a ions, and Ou comes
The e alua ion s udy was an essen ial s ep in alida ing he concep s behind MuSA.
Despi e i s limi a ions—a small sample size wi h ew usable da a, he use o di e -
en mic ophones, and he emo e, asynch onous na u e o he s udy—i p oduced
a numbe o indings. Subjec i e eedback om he pos -s udy ques ionnai e indi-
ca ed ha pa icipan s pe cei ed MuSA as a po en ially help ul ool, pa icula ly
o imp o ing awa eness o dynamics. This aligns wi h he o e all posi i e sen imen
exp essed in he sen imen analysis, whe e pa icipan s app ecia ed he cla i y and
use ulness o highligh ed salien momen s. Mino usabili y issues we e no ed, in-
cluding lack o eal- ime esponsi eness and occasional misalignmen be ween isual
eedback and ac ual pe o mance, highligh ing oppo uni ies o e ining he use
expe ience.
Objec i e pe o mance me ics. The objec i e pe o mance me ics e ealed a
mo e nuanced pic u e o MuSA’s impac ac oss skill le els and musical ea u es. Dy-
namics consis en ly bene i ed he mos om eedback, whe eas pi ch imp o emen s
we e minimal and empo ou comes we e inconsis en . Expe ience le el u he mod-
e a ed hese e ec s: in e media e musicians, co esponding o he Compe ence s age
o he TAD model, o en demons a ed measu able gains, while ad anced musicians,
co esponding o he Expe ise s age, some imes showed neu al o sligh ly nega i e
changes, pa icula ly o empo. Al hough hese ends we e no s a is ically sig-
ni ican , hey poin o he expe ise e e sal e ec , whe eby eedback me hods ha
bene i less expe ienced lea ne s may dis up he pe o mance s a egies o expe s
[194, 195, 196, 197, 198]. This also aligns wi h Colombo (2017), who sugges ed
ha young music s uden s bene i om g ea e eache suppo and moni o ing,
while mo e ad anced s uden s gain mo e om de eloping me acogni i e skills such
as e lec ing on hei lea ning s a egies [199].
Pe cei ed pe o mance and sel -awa eness. Analysis o pe cei ed pe o mance
e ealed ha pa icipan s epo ed he g ea es imp o emen o dynamics, mode -
a e imp o emen o empo, and minimal change o pi ch. No ably, some empo
72 Chap e 5. Discussion
imp o emen s appea ed o esul om p ac ice alone a he han ool-media ed eed-
back, emphasizing he impo ance o dis inguishing in insic lea ning e ec s om he
in luence o MuSA. Sel -assessmen me ics demons a ed gene ally weak alignmen
wi h objec i e pe o mance, sugges ing limi ed sel -awa eness. In e media e musi-
cians showed modes ly be e sel -assessmen accu acy, whe eas ad anced musicians
we e o en in e sely aligned wi h hei measu ed ou comes. These indings align wi h
esea ch on musician sel -awa eness, which sugges s ha lea ne s’ sel -assessmen s
can be inaccu a e [200]. O e all, poo sel -assessmen may di ec ly e lec o a chal-
lenge in he sel - e lec ion phase o he SRL cycle. They also highligh he complex
in e ac ion be ween expe ience, ea u e ype, and sel -e alua ion in music p ac ice,
indica ing ha MuSA could be enhanced wi h me acogni i e p omp s o suppo
mo e accu a e pe o mance judgmen s.
Taken oge he , hese esul s indica e ha MuSA may selec i ely suppo musi-
cal p ac ice as mos e ec i e o dynamics and o lea ne s a in e media e skill
le els, while ad anced lea ne s may expe ience neu al o sligh ly nega i e e ec s.
Subjec i e imp essions sugges ha he ool success ully di ec s a en ion o salien
pe o mance momen s, al hough u u e i e a ions could bene i om imp o emen s
in in e ac i i y and alignmen o isual eedback wi h audio. These insigh s unde -
sco e he impo ance o combining subjec i e and objec i e measu es when e alua -
ing digi al eedback ools and poin owa d p ac ical design conside a ions, including
ea u e-speci ic guidance and adap i e eedback calib a ed o expe ience le el.
Limi a ions. One o he main limi a ions o he MuSA e alua ion s udy was ha
i es ed he applica ion as a whole, a he han isola ing he e ec o salien momen
highligh ing i sel . Pa icipan s we e asked o pe o m a ask wi h he ull analyze
and wi hou i , meaning ha only gene al s a emen s abou MuSA’s o e all u ili y
can be made, a he han conclusions speci ically abou he impac o indica ing
salien momen s e sus displaying aw ea u e cu es alone. This dis inc ion is
impo an , pa icula ly because empo saliency analysis was no implemen ed due o
algo i hmic limi a ions. In e es ingly, his p o ides a na u al compa ison: dynamics
eedback showed a mo e consis en ly posi i e end on pe o mance, whe eas empo
5.1. Discussion 73
eedback—absen in he inal e sion—migh ha e had a ma kedly nega i e e ec i
implemen ed, illus a ing a po en ial di e en ial impac o ea u e ype on p ac ice
ou comes.
Ano he limi a ion is ha he s udy did no speci ically assess whe he MuSA im-
p o es a musician’s exp essi e in en o abili y o c ea e exp essi e pe o mance mo-
men s, and mo eo e , didn’ explici ly con i m i s use as a e lec i e, non-co ec i e,
pos -pe o mance ool. Tes ing his unc ionali y and acqui ing mo e quali a i e
da a ega ding i s e lec i e po en ial would ha e equi ed ec ui ing bo h highly
skilled musicians capable o in en ionally manipula ing exp essi e pa ame e s and
less expe ienced pa icipan s, as well as acili a ing ex ended, ca e ully s uc u ed
p ac ice sessions—cons ain s ha exceeded he a ailable hesis imeline, especially
gi en he ime-in ensi e p ocess o de eloping he MuSA web applica ion i sel .
Addi ionally, he choice o wo simple pieces based on melodic segmen s om com-
mon child en’s songs may no ha e been adequa e o e alua e he use o MuSA ac oss
all s ages o he TAD Music Model. Fo example, he simplici y o he e e ence au-
dios may ha e been i ial o ad anced pa icipan s, also po en ially explaining he
in e se ela ionship be ween expe ise le el and imp o emen . Ideally, he e alu-
a ion would be using di e en music pieces ele an o each pa icipan , bu his
would ha e in oduced an ex a a iable, pe haps o e complica ing he e alua ion
s udy based on he gi en hesis ime ame.
Finally, he small usable sample size (14 ou o 32 pa icipan s) limi s he gene -
alizabili y o he indings. Mos obse ed e ec s we e no s a is ically signi ican ,
making i di icul o d aw de ini i e conclusions. Ne e heless, he s udy e ealed
in iguing pa e ns ega ding ea u e-speci ic and expe ience-dependen e ec s o
eedback, p o iding a aluable benchma k and se ing he s age o u u e esea ch
and i e a ions o MuSA o simila non- eal- ime eedback applica ions.
Implica ions. Ne e heless, he eedback om pa icipan s and he ini ial eques
om RCM musicians highligh a clea , unme need o non- eal- ime eedback ools
ha engage wi h p ac ice sel - e lec ion. Exis ing ools, such as Sonic Analyse and
74 Chap e 5. Discussion
Audaci y, a e powe ul bu lack pedagogical o ien a ion and he abili y o abs ac
aw audio in o meaning ul, musically salien momen s. MuSA, he e o e, ep esen s
a p omising app oach, bu u he esea ch wi h a la ge , mo e di e se pa icipan
sample is needed o alida e i s pedagogical impac and e ine which ypes o salien
momen analysis a e mos e ec i e ac oss skill le els.
5.1.3 Independen P ac ice and Lea ne Implica ions
De eloping MuSA was an impo an s ep in explo ing how echnology can enhance
independen p ac ice, which is o en uns uc u ed and lacks expe eedback. As dis-
cussed in Sec ion 2.2.5, independen p ac ice equi es sel - egula ed lea ning (SRL)
skills, such as sel - e lec ion and goal-se ing, which a e challenging o lea ne s wi h-
ou consis en eache guidance. By highligh ing salien momen s, MuSA aims o
sca old he sel -moni o ing p ocess, di ec ing a lea ne ’s a en ion o speci ic poin s
o e lec ion and expe imen a ion. This suppo s a cycle o e lec i e lea ning ha
is c ucial o de eloping sel -di ec ed lea ning skills.
This po en ial bene i is pa icula ly impo an gi en he challenges lea ne s ace
wi h me acogni ion, o he awa eness o hei own hinking and lea ning p ocesses.
The sel -awa eness me ics om his s udy sugges ha me acogni i e accu acy is a
signi ican hu dle in independen p ac ice, as pa icipan s ac oss he boa d showed
gene ally poo alignmen be ween hei subjec i e pe cep ions and objec i e pe o -
mance ou comes ( =−0.122). This inding aligns wi h b oade esea ch indica ing
ha lea ne s o en s uggle o accu a ely assess hei own pe o mance [201, 202].
In e es ingly, while he sel -e alua ion s udy sugges ed ha in e media e (Compe-
ence) musicians we e mode a ely accu a e in judging hei own pe o mances a e
using MuSA, his pa e n may pa ly e lec sel -selec ion bias: indi iduals iden i-
ying as in e media e may app oach sel -assessmen and e lec ion mo e cau iously
and analy ically, whe eas hose iden i ying as ad anced (Expe ise) may ely mo e
on in e nalized, emo ionally d i en expec a ions o how hey should pe o m a hei
le el.
Ne e heless, his o e all lack o sel -awa eness unde sco es he impo ance o ex-
5.1. Discussion 75
e nal ools in suppo ing independen lea ning. Since me acogni i e skills a e no
always inna e, echnology can p o ide a conc e e ounda ion o e lec ion, whe he
h ough bina y co ec /inco ec eedback ha highligh s salien momen s (as MuSA
does) o by iden i ying pe o mance cha ac e is ics such as a ec . Wi h espec o
imp o ing pe o mance sel -awa eness (which was no MuSA’s p ima y goal), he
p esen s udy did no e eal clea gains om using he analyze ool, bu i is di -
icul o d aw de ini i e conclusions. Sel -awa eness ypically de elops g adually
wi hin sca olding amewo ks and dialogic lea ning con ex s, whe e sus ained ex-
e nal eedback such as guidance om a u o suppo s i s g ow h o e ime. I
is possible ha ex ended use o a ool like MuSA could con ibu e o shaping sel -
awa eness, pa icula ly as s udy pa icipan s we e also managing he cogni i e load
o lea ning how o use he sys em, which may ha e in luenced sel -awa eness esul s.
5.1.4 MuSA’s Limi a ions and Fu u e Di ec ions
Expanding ea u e-d i en saliency analyses. MuSA’s cu en app oach o
saliency analysis is p ima ily ea u e-d i en, ocusing on iden i ying momen s whe e
objec i e musical ea u es—such as pi ch, dynamics, o empo—show signi ican
a iabili y ela i e o an es ablished pe o mance no m. Du ing he implemen a ion
phase, howe e , se e al oadblocks eme ged. Fo example, highligh ing momen s
o high empo a iabili y p o ed di icul because he amoun o empo da a ex-
ac ed om a piece was much lowe compa ed o pi ch o dynamics. In addi ion,
de eloping a signal p ocessing algo i hm ha could consis en ly handle sho e au-
dio segmen s wi hin he gi en ime cons ain s was no easible. As a esul , empo
a iabili y could no be highligh ed as eliably as o he ea u es. Fu u e wo k should
he e o e p io i ize implemen ing mo e obus algo i hms o empo analysis, as he
MuSA amewo k al eady suppo s such unc ionali y and only equi es he igh
compu a ional app oach.
Beyond empo, o he ea u es p esen oppo uni ies o u he e inemen . Fo in-
s ance, MuSA cu en ly iden i ies ib a o momen s using a ixed h eshold. While
his me hod wo ks, i misses he exp essi e nuance con eyed h ough changes in
76 Chap e 5. Discussion
ib a o a e and ex en . Yang e al. (2013) demons a ed ha a iabili y in i-
b a o can signal exp essi e in en [25], sugges ing ha saliency de ec ion could be
imp o ed by acking ib a o a iabili y ac oss a pe o mance a he han applying
a ha d cu o . This would allow he ool o cap u e mo e abs ac and musically
meaning ul pe o mance momen s.
Simila ly, phona ion mode has been implemen ed o oice, bu i s ole in saliency
emains an open ques ion. Is i he ansi ion om one phona ion mode o ano he
ha c ea es salience? O is i sus ained use o a pa icula mode in con as o su -
ounding ma e ial? In es iga ing such ques ions would no only e ine how salience
is ope a ionalized bu also expand he in e p e i e powe o MuSA in analyzing
exp essi e ocal pe o mance.
In eg a ing mo e ea u es. A missed oppo uni y in MuSA was also o de ec
salien momen s h ough imb e, o example, as imb e is one o he p ima y mech-
anisms h ough which we de ec emo ion in music [43, 24]. This could, o example,
in ol e de ec ing and displaying a iable momen s in imb e ela ed low-le el ea-
u es like spec al cen oid and lux o use s and le ing hem in e p e musical
meaning and unde s anding om hese highligh ed a eas based on hei in en ion.
Though he o iginal p o o ype o MuSA included highligh ing he mos a iable
imb e sec ion based on MFCCs, he inal e sion did no include i o iming and
scheduling easons based on wo kload and he design o he applica ion i sel .
In eg a ing a ec -d i en saliency analysis. Mo eo e , a complemen a y ap-
p oach o he ea u e-based one could in ol e a ec -d i en saliency analysis, such
as ac i i y analysis [180]. In his case, a model could be ained o iden i y key mo-
men s ac oss di e en ins umen al pe o mances—s a ing, o ins ance, wi h oice
and iolin—by de ec ing poin s whe e lis ene s consis en ly pe cei e a pa icula
a ec i e quali y, such as sadness o happiness.
In eg a ing ac i i y analysis would also align mo e explici ly wi h he TDPK Music
F amewo k (Sec ion 2.4), which emphasizes ha musical expe iences a e insepa a-
ble om hei exp essi e cha ac e . Compa ed o ea u e-d i en saliency analysis,
5.1. Discussion 77
his app oach would p o ide mo e p esc ip i e eedback abou po en ial a ec and
musical exp ession. A he same ime, as he e alua ion s udy sugges s, such p e-
sc ip i e eedback on exp essi e sec ions migh be mos e ec i e o lea ne s who a e
s ill de eloping hese skills, while mo e ad anced musicians may bene i less om
his ype o guidance (a po en ial e e se expe e ec as p e iously men ioned).
Need o musical unde s anding. This poin s o ano he po en ial limi a ion
o he ool: e ec i ely in e p e ing he eedback p o ided by MuSA likely equi es a
ce ain le el o musical unde s anding, ypically a he in e media e le el o beyond.
Fo beginne s, an explici ly p esc ip i e, ask-le el eedback ool may be mo e e -
ec i e, as sugges ed by he McPhe son e al. eedback ma ix [130]. A he same
ime, eedback mus be app op ia ely ailo ed so as no o in e e e wi h he sel -
egula o y abili ies ha in e media e and ad anced lea ne s ha e de eloped. In his
sense, he ac ha MuSA is no o e ly p esc ip i e and ins ead equi es some in-
e p e i e backg ound knowledge may, om he lea ne ’s pe spec i e, be conside ed
a bene i a he han a d awback.
Musa as a web applica ion. One explici bene i o MuSA is i s ully online
o ma . Use s do no need o download so wa e, c ea e an accoun , o na iga e
complex se up p ocesses. The ange o a ailable analyses is in en ionally limi ed,
which some migh iew as a d awback since use s canno access an ex ensi e menu
o op ions o e iew pas uploads. Howe e , his simplici y also educes ba ie s
o use: lea ne s can ei he eco d di ec ly wi hin he applica ion o upload exis -
ing audio iles and ecei e isual eedback immedia ely. While ools such as Sonic
Visualise o e sophis ica ed analysis capabili ies, hey we e no designed wi h a
pedagogical audience in mind, bu a he o compu a ional esea ch, and a e he e-
o e unde u ilized in pe o mance e iew con ex s. By con as , MuSA’s s eamlined
design makes i mo e accessible o musicians, minimizing cogni i e load compa ed
o he s eep lea ning cu e o downloading, con igu ing, and in e p e ing esul s
om specialized so wa e like Sonic Visualise .
78 Chap e 5. Discussion
Cu en a chi ec u al limi a ions. Cu en ly, MuSA unc ions as a minimum
iable p oduc . While i can handle a la ge numbe o simul aneous use in e -
ac ions h ough i s Redis-based se up, i is no ye op imized o la ge-scale de-
ploymen . Cu en cons ain s include sha ed se e esou ces, edundancy om
he dual Node.js and Flask se ices—which could be s eamlined by consolida ing
da abase calls in o he Py hon se ice—and eliance on local ile s o age, which
would bene i om a cloud-based solu ion o imp o ed scalabili y and esilience.
All collec ed da a is s o ed di ec ly on he se e , which has limi ed capaci y. While
his se up allows da a o be e ained, i also in oduces he challenge o managing
and main aining g owing s o age demands. In p ac ice, his means ha someone
mus pe iodically ans e he collec ed audio iles o an ex e nal d i e, a p ocess
ha can be bo h ime-consuming and incon enien . Addi ionally, he de elopmen
s ack uses C ea e Reac App, which has been dep eca ed; mig a ing o Vi e would
mode nize he build p ocess and imp o e e iciency. Add essing hese edundancies
and in as uc u e limi a ions is a p io i y o u u e e sions, bu due o ime con-
s ain s, a ully comme cial-le el ool was no easible. Ne e heless, MuSA p o ides
a unc ional ounda ion o deli e ing salience-based analysis o lea ne s, enabling
non- eal- ime pe o mance eedback ha can in o m u u e i e a ions and ela ed
p ojec s.
In eg a ing w i en no a ion. In e ms in eg a ion wi h no a ion, MuSA does
no cu en ly employ sco e alignmen wi h pe o mances, which some may iew as
a limi a ion. Sco e alignmen , howe e , comes wi h i s own d awback: i equi es
he pe o me o ha e a no a ed sco e o compa ison, which may no always be
a ailable o ele an . MuSA delibe a ely a oids his ea u e because i s ocus is on
he pe o mance i sel , p i ileging he pe o me ’s exp essi e in en o e adhe ence
o no a ion, and dis ancing i sel om he bina y co ec /inco ec p esc ip ions
seen in eal- ime eedback ools like Yousician. Tha said, sco e alignmen can also
o e aluable insigh s in o pe o mance and exp essi e in en , and se e al s udies
highligh i s use ulness. Fu u e i e a ions o MuSA could he e o e in eg a e sco e
alignmen as an op ional ea u e o enhance salien -momen analysis, pa icula ly
5.1. Discussion 79
when key sco e e en s a e in ol ed. Such in eg a ion could e en allow use s o se
a iabili y a ge s based on he sco e o ecei e eedback amed in mo e adi ional
musical e ms. Impo an ly, inco po a ing sco e alignmen may be mo e easible in
MuSA’s non– eal- ime se ing, since eal- ime sco e ollowing emains echnically
challenging o implemen eliably.
Pe o mance- o-pe o mance alignmen . Ano he ea u e o iginally in ended
o MuSA bu no implemen ed due o ime and esou ce cons ain s was di ec
compa ison o a a ge audio eco ding. This would ha e been aluable because he
e e ence pe o mance i sel con ains ea u e-d i en salien momen s ha could be
ex ac ed and di ec ly compa ed o hose in he use ’s eco ding. While compa a-
i e pe o mance analysis is al eady used in he ield, he e emains a need o mo e
sophis ica ed app oaches beyond simple e o de ec ion [203]. In his sense, MuSA
could e en ually be expanded o compa e salien momen s be ween a s uden ’s pe -
o mance and a e e ence eco ding, o e ing eedback ha is non-bina y and mo e
o ien ed owa d exp essi e in en and musicali y. Howe e , implemen ing such a
ea u e was beyond he scope o his i e a ion, especially gi en he echnical and
esou ce demands o building a la ge-scale, mul i-use applica ion.
Da a collec ion po en ial. While MuSA was p ima ily designed as a pedagog-
ical applica ion, i also unc ions as a aluable da a collec ion ool. Since MuSA is
deployed on a se e hos ed by he MTG a Uni e si a Pompeu Fab a, all uploaded
eco dings a e au oma ically s o ed (wi h use s explici ly consen ing o da a collec-
ion be o e using he ool). Addi ionally, each eco ding is labeled by ins umen
(cu en ly oice and iolin), allowing MuSA o se e as a passi e, labeled da ase
o hese ins umen s. These eco dings, accessible o hose wi h se e pe missions,
could be epu posed o u u e music pe o mance analysis o music in o ma ion
e ie al asks equi ing labeled ins umen al da a.
Lis o Tables
1 Fea u e ex ac ion and salien momen selec ion me hods (MuSA V1). 37
2 Fea u e ex ac ion me hods and c i e ia o salien momen selec ion
(MuSA V2) wi h exac pa ame e s. . . . . . . . . . . . . . . . . . . . 39
3 Ques ionnai e i ems and co esponding labels. . . . . . . . . . . . . . 66
4 Two-way ANOVA esul s o Raw RMSE pe o mance (Condi ion ×
Timing)...................................114
5 G oup means and e ec sizes (Cohen’s d) o RMSE pe o mance.
Signi icance: ** p < 0.01. . . . . . . . . . . . . . . . . . . . . . . . . 114
6 Pi ch (Hz) RMSE compa isons by expe ience le el. . . . . . . . . . . 115
7 Dynamics (RMS) RMSE compa isons by expe ience le el. . . . . . . . 115
8 Tempo (BPM) RMSE compa isons by expe ience le el. . . . . . . . . 115
9 O e all baseline-co ec ed RMSE imp o emen s compa ing eedback
and con ol condi ions. Cohen’s dand p- alues indica e nonsigni ican
di e ences..................................116
10 Baseline-co ec ed RMSE imp o emen s (%) o pi ch, dynamics, and
empo by musician expe ience le el. Posi i e alues indica e an ad-
an age o he eedback condi ion; all compa isons we e nonsigni ican .116
11 O e all Z-Sco e S anda dized Imp o emen : Con ol s Feedback . . 116
12 Z-Sco e S anda dized Imp o emen by Expe ience Le el: Con ol s
Feedback..................................116
13 O e all Pe cei ed Pe o mance Ra ings by Condi ion. Signi icance: *
p<0.05,***p<0.001..........................116
14 Pe cei ed Pe o mance Imp o emen (∆= A e - Be o e) wi h -
s a is ics and p- alues by Expe ience Le el and Condi ion. . . . . . . 117
86
LIST OF TABLES 87
15 Co ela ion Analysis o Sel -Assessmen Accu acy by Fea u e . . . . . 117
16 Sel -Awa eness Compa ison: Con ol s Feedback Condi ions . . . . . 117
17 Sel -Awa eness by Musical Expe ience Le el (Co ela ion wi h Objec-
i ePe o mance).............................117
18 Rep esen a i e pa icipan quo es by ques ion and sen imen . . . . . . 118
19 Summa y o sen imen analysis esul s, including o e all s a is ics,
pe -ques ion means and medians, and co ela ion be ween lexicon and
VADERsen imen .............................119
Bibliog aphy
[1] Technology enhanced lea ning o musical ins umen pe o mance ( elmi). Eu-
opean Commission Ho izon 2020 Resea ch and Inno a ion P og amme, G an
Ag eemen No. 688269 (2016–2019). URL h ps:// elmi.up .edu/. P ojec
du a ion: Feb ua y 2016 – Janua y 2019.
[2] E icsson, K. A. The Scien i ic S udy o Expe Le els o Pe o mance: gene al
implica ions o op imal lea ning and c ea i i y 1. High Abili y S udies 9, 75–
100 (1998). URL h ps://doi.o g/10.1080/1359813980090106. Publishe :
Rou ledge.
[3] McPhe son, G. E., Blackwell, J. & Hallam, S. Musical Po en ial, Gi edness,
and Talen De elopmen . In McPhe son, G. E. (ed.) The Ox o d Handbook o
Music Pe o mance, Volume 1, 31–55 (Ox o d Uni e si y P ess, 2022). URL
h ps://doi.o g/10.1093/ox o dhb/9780190056285.013.3.
[4] Pla z, F., Kopiez, R., Lehmann, A. C. & Wol , A. The in luence o delibe a e
p ac ice on musical achie emen : a me a-analysis. F on ie s in Psychology 5
(2014).
[5] Macnama a, B. N., Hamb ick, D. Z. & Oswald, F. L. Delibe a e P ac ice
and Pe o mance in Music, Games, Spo s, Educa ion, and P o essions: A
Me a-Analysis. Psychological Science 25, 1608–1618 (2014). URL h ps:
//doi.o g/10.1177/0956797614535810. Publishe : SAGE Publica ions Inc.
88
BIBLIOGRAPHY 89
[6] E icsson, K. A., K ampe, R. T. & Tesch-Röme , C. The ole o delibe a e
p ac ice in he acquisi ion o expe pe o mance. Psychological Re iew 100,
363–406 (1993). Place: US Publishe : Ame ican Psychological Associa ion.
[7] Meinz, E. J. & Hamb ick, D. Z. Delibe a e p ac ice is necessa y bu no
su icien o explain indi idual di e ences in piano sigh - eading skill: The
ole o wo king memo y capaci y. Psychological Science 21, 914–919 (2010).
Place: US Publishe : Sage Publica ions.
[8] Hallam, S. e al. The de elopmen o p ac ising s a egies in young people.
Psychology o Music 40, 652–680 (2012). URL h ps://doi.o g/10.1177/
0305735612443868. Publishe : SAGE Publica ions L d.
[9] McPhe son, G. E., & Renwick, J. M. A Longi udinal S udy o Sel - egula ion
in Child en’s Musical P ac ice. Music Educa ion Resea ch 3, 169–186 (2001).
URL h ps://doi.o g/10.1080/14613800120089232. Publishe : Rou -
ledge.
[10] Gaun , H., , C., And ea, , L., Ma ion, & Hallam, S. Suppo ing conse a oi e
s uden s owa ds p o essional in eg a ion: one- o-one ui ion and he po en ial
o men o ing. Music Educa ion Resea ch 14, 25–43 (2012). URL h ps:
//doi.o g/10.1080/14613808.2012.657166. Publishe : Rou ledge.
[11] Gaun , H. App en iceship and empowe men : he ole o one- o-one lessons. In
Rink, J., Gaun , H. & Williamon, A. (eds.) Musicians in he Making: Pa hways
o C ea i e Pe o mance, 0 (Ox o d Uni e si y P ess, 2017). URL h ps:
//doi.o g/10.1093/acp o :oso/9780199346677.003.0003.
[12] Phillips, J. A. S uden sel -assessmen and e lec ion in a lea ne con olled
en i onmen (2016). URL h ps://a xi .o g/abs/1608.00313.
[13] Dos San os Sil a, C. & Ma inho, H. Sel - egula ed lea ning p ocesses o ad-
anced musicians: A p isma e iew. Musicae Scien iae 29, 89–108 (2025).
URL h ps://jou nals.sagepub.com/doi/10.1177/10298649241275614.
90 BIBLIOGRAPHY
[14] Magoon, A. J. Cons uc i is app oaches in educa ional esea ch. Re iew o
Educa ional Resea ch 47, 651–693 (1977). URL h ps://jou nals.sagepub.
com/doi/10.3102/00346543047004651.
[15] Lim, C. P. & Chan, B. C. mic olessons in eache educa ion: Examining
p e-se ice eache s’ pedagogical belie s. Compu e s Educa ion 48, 474–
494 (2007). URL h ps://www.sciencedi ec .com/science/a icle/pii/
S0360131505000424.
[16] Ha ie, J. Visible Lea ning o Teache s (Rou ledge, 2012), 0 edn. URL
h ps://www. aylo ancis.com/books/9781136592331.
[17] Angeli, C. & Valanides, N. P ese ice elemen a y eache s as in o ma ion and
communica ion echnology designe s: An ins uc ional sys ems design model
based on an expanded iew o pedagogical con en knowledge. Jou nal o
compu e assis ed lea ning 21, 292–302 (2005).
[18] Angeli, C. & Valanides, N. Epis emological and me hodological issues o
he concep ualiza ion, de elopmen , and assessmen o ic – pck: Ad ances in
echnological pedagogical con en knowledge ( pck). Compu e s & educa ion
52, 154–168 (2009).
[19] Angeli, C. & Valanides, N. Technology mapping: An app oach o de el-
oping echnological pedagogical con en knowledge. Jou nal o Educa ional
Compu ing Resea ch 48, 199–221 (2013).
[20] Ha ie, J. & Timpe ley, H. The powe o eedback. Re iew o educa ional
esea ch 77, 81–112 (2007).
[21] Sloboda, J. A. The ole o musical s uc u e in emo ional esponse. Psychology
o Music 24, 34–50 (1996).
[22] Juslin, P. N. Cue u iliza ion in communica ion o emo ion in music pe o -
mance: Rela ing exp essi e in en ion o acous ic ea u es. Jou nal o Ex-
pe imen al Psychology: Human Pe cep ion and Pe o mance 26, 1797–1813
(2000).
BIBLIOGRAPHY 91
[23] Gómez, E. & He e a, P. An app oach o he analysis o musical exp essi e
iming in piano pe o mance eco dings. Jou nal o New Music Resea ch 35,
3–17 (2006).
[24] Ee ola, T. & e al. An op imized ac o ial design o s udying he emo ion
pe cep ion om music. In In e na ional Con e ence on Music Pe cep ion and
Cogni ion (2012).
[25] Yang, L. & e al. Vib a o pe o mance s yle: a case s udy compa ing e hu and
iolin. In 10 h In e na ional Symposium on Compu e Music Modeling and
Re ie al (CMMR 2013), 109–116 (CMMR, 2013).
[26] Li, P.-C., Su, L., Yang, Y.-H. & Su, A. W. Y. Analysis o exp essi e musical
e ms in iolin using sco e-in o med and exp ession-based audio ea u es. In
P oceedings o he 16 h In e na ional Socie y o Music In o ma ion Re ie al
Con e ence, 441–448 (ISMIR, 2015).
[27] Baue , W. I., Reese, S. & McAllis e , P. A. T ans o ming music eaching
ia echnology: The ole o p o essional de elopmen . Jou nal o Resea ch
in Music Educa ion 51, 289–301 (2003). URL h ps://jou nals.sagepub.
com/doi/10.2307/3345656.
[28] Sa age, J. A su ey o ic usage ac oss english seconda y schools. Music
Educa ion Resea ch 12, 89–104 (2010). URL h p://www. and online.com/
doi/abs/10.1080/14613800903568288.
[29] G ehe , G. R. Music echnology pa ne ships: A con ex o music eache
p epa a ion. A s Educa ion Policy Re iew 112, 130–136 (2011). URL h p:
//www. and online.com/doi/abs/10.1080/10632913.2011.566083.
[30] Wise, S., G eenwood, J. & Da is, N. Teache s’ use o digi al echnology in sec-
onda y music educa ion: illus a ions o changing class ooms. B i ish Jou nal
o Music Educa ion 28, 117–134 (2011). URL h ps://www.camb idge.o g/
co e/p oduc /iden i ie /S0265051711000039/ ype/jou nal_a icle.
92 BIBLIOGRAPHY
[31] M oziak, J. & Bowman, J. Music pack in highe educa ion: Educa ing he
educa o s. Handbook o echnological pedagogical con en knowledge (TPACK)
o educa o s 285–296 (2016).
[32] Miksza, P. Rela ionships among achie emen goal mo i a ion, impulsi i y, and
he music p ac ice o collegia e b ass and woodwind playe s. Psychology o Mu-
sic 39, 50–67 (2011). URL h ps://doi.o g/10.1177/0305735610361996.
Publishe : SAGE Publica ions L d.
[33] Miksza, P., Blackwell, J. & Rose h, N. E. Sel -Regula ed Music P ac-
ice. Jou nal o Resea ch in Music Educa ion 66, 295–319 (2018). URL
h ps://www.js o .o g/s able/48588925. Publishe : [MENC: The Na-
ional Associa ion o Music Educa ion, Sage Publica ions, Inc.].
[34] Acquilino, A. & Sca one, G. Cu en s a e and u u e di ec ions o ech-
nologies o music ins umen pedagogy. F on ie s in Psychology 13, 835609
(2022). URL h ps://www. on ie sin.o g/a icles/10.3389/ psyg.
2022.835609/ ull.
[35] Tho ndike, E. Educa ional Psychology. Vol. I: The O iginal Na u e o Man.
Educa ional Psychology. Vol. I: The O iginal Na u e o Man. (New Yo k,
Columbia Uni ., 1913). Pages: Pp. xii+, 327.
[36] Council, N. R. e al. How people lea n: B ain, mind, expe ience, and school:
Expanded edi ion, ol. 1 (Na ional Academies P ess, 2000).
[37] Swelle , J. Cogni i e load heo y, lea ning di icul y, and ins uc ional design.
Lea ning and ins uc ion 4, 295–312 (1994).
[38] Swelle , J. Cogni i e load heo y. The psychology o lea ning and mo i a ion:
Cogni ion in educa ion, Vol. 55 37–76 (2011). Place: San Diego, CA, US
Publishe : Else ie Academic P ess.
[39] Rego, C. & Mon ague, E. The impac o eedback modali ies and he in lu-
ence o cogni i e load on in e pe sonal communica ion in nonclinical se ings:
BIBLIOGRAPHY 93
Expe imen al s udy design. JMIR Human Fac o s 10, e49675 (2023). URL
h ps://human ac o s.jmi .o g/2023/1/e49675.
[40] Lu ie, N. H. & Swamina han, J. M. Is imely in o ma ion always be e ?
he e ec o eedback equency on decision making. O ganiza ional Beha io
and Human Decision P ocesses 108, 315–329 (2009). URL h ps://www.
sciencedi ec .com/science/a icle/pii/S0749597808000745.
[41] Ka lsson, J. & Juslin, P. N. Musical exp ession: an obse a ional s udy o
ins umen al eaching. Psychology o Music 36, 309–334 (2008). URL h ps:
//jou nals.sagepub.com/doi/10.1177/0305735607086040.
[42] McPhe son, G. E., Da idson, J. W. & Faulkne , R. Music In Ou Li es: Re-
hinking Musical Abili y, De elopmen , and Iden i y (Ox o d Uni e si y P ess,
2012). URL h ps://doi.o g/10.1093/acp o :oso/9780199579297.001.
0001.
[43] Gab ielsson, A. Music Pe o mance Resea ch a he Millennium. Psy-
chology o Music 31, 221–272 (2003). URL h ps://doi.o g/10.1177/
03057356030313002. Publishe : SAGE Publica ions L d.
[44] P eckel, F. e al. Talen de elopmen in achie emen domains: A psy-
chological amewo k o wi hin- and c oss-domain esea ch. Pe spec i es
on Psychological Science 15, 691–722 (2020). URL h ps://doi.o g/10.
1177/1745691619895030. PMID: 32196409, h ps://doi.o g/10.1177/
1745691619895030.
[45] Müllensie en, D. e al. 84 alen de elopmen in music. In The Ox o d Hand-
book o Music Pe o mance, Volume 1 (Ox o d Uni e si y P ess, 2022). URL
h ps://doi.o g/10.1093/ox o dhb/9780190056285.013.32.h ps:
//academic.oup.com/book/0/chap e /357714233/chap e -ag-pd /
57022543/book_42624_sec ion_357714233.ag.pd .
[46] Johns on, H. The use o ideo sel -assessmen , pee -assessmen ,
and ins uc o eedback in e alua ing conduc ing skills in music s u-
94 BIBLIOGRAPHY
den eache s. B i ish Jou nal o Music Educa ion 10, 57–63
(1993). URL h ps://www.camb idge.o g/co e/p oduc /iden i ie /
S0265051700001431/ ype/jou nal_a icle.
[47] Robinson, C. R. Singe s’ sel -assessmen o cho al pe o mance: nex -day
ecollec ions e sus conce ape e alua ion. Sou heas e n J. Music Educ 4,
224–233 (1993).
[48] Daniel, R. Sel -assessmen in pe o mance. B i ish Jou nal o Music Educa-
ion 18, 215–226 (2001). URL h ps://www.camb idge.o g/co e/p oduc /
iden i ie /S0265051701000316/ ype/jou nal_a icle.
[49] Hewi , M. P. Sel -e alua ion endencies o junio high ins umen alis s.
Jou nal o Resea ch in Music Educa ion 50, 215–226 (2002). URL h ps:
//jou nals.sagepub.com/doi/10.2307/3345799.
[50] Sil ei a, J. M. & Ga in, R. The e ec o audio eco ding and playback
on sel -assessmen among middle school ins umen al music s uden s. Psy-
chology o Music 44, 880–892 (2016). URL h ps://doi.o g/10.1177/
0305735615596375. Publishe : SAGE Publica ions L d.
[51] Woody, R. H. Lea ning exp essi i y in music pe o mance: An explo a o y
s udy. Resea ch S udies in Music Educa ion 14, 14–23 (2000). URL h ps:
//jou nals.sagepub.com/doi/10.1177/1321103X0001400102.
[52] Woody, R. H. Lea ning om he expe s: Applying esea ch in expe pe -
o mance o music educa ion. Upda e: Applica ions o Resea ch in Music
Educa ion 19, 9–14 (2001). URL h ps://jou nals.sagepub.com/doi/10.
1177/87551233010190020103.
[53] Williamon, A. Musical ExcellenceS a egies and Techniques o
Enhance Pe o mance (Ox o d Uni e si y P ess, 2004). URL
h p://www.ox o dschola ship.com/ iew/10.1093/acp o :oso/
9780198525356.001.0001/acp o -9780198525356.
BIBLIOGRAPHY 95
[54] Juslin, P. N., Ka lsson, J., Linds öm, E., F ibe g, A. & Schoonde wald , E.
Play i again wi h eeling: Compu e eedback in musical communica ion o
emo ions. Jou nal o Expe imen al Psychology: Applied 12, 79–95 (2006).
URL h ps://doi.apa.o g/doi/10.1037/1076-898X.12.2.79.
[55] Timme s, R. & Sadaka a, M. T aining Exp essi e Pe o mance by Means o
Visual Feedback: Exis ing and Po en ial Applica ions o Pe o mance Mea-
su emen Techniques, 304–327 (Ox o d Uni e si y P essOx o d, 2014), 1 edn.
URL h ps://academic.oup.com/book/27746/chap e /197952585.
[56] Zimme man, B. J. A social cogni i e iew o sel - egula ed academic lea ning.
Jou nal o educa ional psychology 81, 329 (1989).
[57] Zimme man, B. J. A aining sel - egula ion: A social cogni i e pe spec i e.
In Handbook o sel - egula ion, 13–39 (Else ie , 2000).
[58] Zimme man, B. J. & Moylan, A. R. Sel - egula ion: Whe e me acogni ion
and mo i a ion in e sec . In Handbook o me acogni ion in educa ion, 299–315
(Rou ledge, 2009).
[59] Ha ield, J. L., Hal a i, H. & Lemy e, P.-N. Ins umen al p ac ice in he
con empo a y music academy: A h ee-phase cycle o sel - egula ed lea ning
in music s uden s. Musicae Scien iae 21, 316–337 (2017). URL h ps://
jou nals.sagepub.com/doi/10.1177/1029864916658342.
[60] Clea y, T. J. & Callan, G. L. Assessing Sel -Regula ed Lea n-
ing Using Mic oanaly ic Me hods, 338–351 (Rou ledge, 2017), 2 edn.
URL h ps://www. aylo ancis.com/books/9781317448662/chap e s/
10.4324/9781315697048-22.
[61] Nusseck, M., Wild, F., Sischka, C. & Spahn, C. E ec s o audio eedback
in e en ions wi h he diskla ie on he pe o mance o piano s uden s. F on-
ie s in Psychology 16, 1568021 (2025). URL h ps://www. on ie sin.
o g/a icles/10.3389/ psyg.2025.1568021/ ull.
102 BIBLIOGRAPHY
[110] Chung *, J. C. C. & Chow, S. M. K. P omo ing s uden lea ning h ough
a s uden -cen ed p oblem-based lea ning subjec cu iculum. Inno a ions
in Educa ion and Teaching In e na ional 41, 157–168 (2004). URL h p:
//www. and online.com/doi/abs/10.1080/1470329042000208684.
[111] Cla k, K. The e ec s o he lipped model o ins uc ion on s uden engagemen
and pe o mance in he seconda y ma hema ics class oom. The Jou nal o
Educa o s Online 12 (2015). URL h ps://www. hejeo.com/a chi e/2015_
12_1/cla k.
[112] Shi ely, J. Cons uc i ism in music educa ion. A s Educa ion Policy Re-
iew 116, 128–136 (2015). URL h p://www. and online.com/doi/ ull/
10.1080/10632913.2015.1011815.
[113] Meissne , H. & Timme s, R. Young musicians’ lea ning o exp essi e pe -
o mance: The impo ance o dialogic eaching and modeling. F on ie s in
Educa ion 5, 11 (2020). URL h ps://www. on ie sin.o g/a icle/10.
3389/ educ.2020.00011/ ull.
[114] Alexande , R. J. Towa ds dialogic eaching: Re hinking class oom alk (2008).
[115] Alexande , R. A dialogic eaching companion (Rou ledge, 2020).
[116] G ano , N., Fische , K. W. & Pa ziale, J. B idging o he unknown: A an-
si ion mechanism in lea ning and de elopmen . Mic ode elopmen : T ansi ion
p ocesses in de elopmen and lea ning 131–156 (2002).
[117] Lajoie, S. P. Ex ending he sca olding me apho . Ins uc ional science 33,
541–557 (2005).
[118] Van de Pol, J., Volman, M. & Beishuizen, J. Sca olding in eache –s uden
in e ac ion: A decade o esea ch. Educa ional psychology e iew 22, 271–296
(2010).
[119] Van De Linden, J., Johnson, R., Bi d, J., Roge s, Y. & Schoonde wald ,
E. Buzzing o play: lessons lea ned om an in he wild s udy o eal- ime
BIBLIOGRAPHY 103
ib o ac ile eedback. In P oceedings o he SIGCHI Con e ence on Human
ac o s in Compu ing Sys ems, 533–542 (2011).
[120] Pea, R. D. The social and echnological dimensions o sca olding and ela ed
heo e ical concep s o lea ning, educa ion, and human ac i i y. Jou nal o
he Lea ning Sciences 13, 423–451 (2004). URL h p://www. and online.
com/doi/abs/10.1207/s15327809jls1303_6.
[121] Reigosa, C. & Jiménez-Aleixand e, M.-P. Sca olded p oblem-sol ing in he
physics and chemis y labo a o y: di icul ies hinde ing s uden s’ assump ion
o esponsibili y. In e na ional Jou nal o Science Educa ion 29, 307–329
(2007).
[122] Küpe s, E., Van Dijk, M., McPhe son, G. & Van Gee , P. A dynamic
model ha links skill acquisi ion wi h sel -de e mina ion in ins umen al mu-
sic lessons. Musicae Scien iae 18, 17–34 (2014). URL h ps://jou nals.
sagepub.com/doi/10.1177/1029864913499181.
[123] Hallam, S. The de elopmen o me acogni ion in musicians: Implica ions o
educa ion. B i ish jou nal o music educa ion 18, 27–39 (2001).
[124] Williamson, S. N. De elopmen o a sel - a ing scale o sel -di ec ed lea ning.
Nu se Resea che 14, 66–83 (2007).
[125] Ga ison, D. R. Sel -Di ec ed Lea ning: Towa d a Comp ehensi e Model.
Adul Educa ion Qua e ly 48, 18–33 (1997). URL h ps://doi.o g/10.
1177/074171369704800103. Publishe : SAGE Publica ions Inc.
[126] Pa kes, K. A. Sel -Di ec ed Lea ning S a egies. In McPhe son, G. E. (ed.)
The Ox o d Handbook o Music Pe o mance, Volume 1, 0 (Ox o d Uni e si y
P ess, 2022). URL h ps://doi.o g/10.1093/ox o dhb/9780190056285.
013.7.
[127] Gibbons, M. The sel -di ec ed lea ning handbook: Challenging adolescen s u-
den s o excel (San F ancisco, 2002).
104 BIBLIOGRAPHY
[128] Rashid, T. & Muhammad Asgha , H. Technology use, sel -di ec ed lea ning,
s uden engagemen and academic pe o mance: Examining he in e ela ions.
Compu e s in Human Beha io 63, 604–612 (2016). Place: Ne he lands Pub-
lishe : Else ie Science.
[129] Sadle , D. R. Fo ma i e assessmen and he design o ins uc ional sys ems.
Ins uc ional Science 18, 119–144 (1989). URL h p://link.sp inge .com/
10.1007/BF00117714.
[130] McPhe son, G. E., Blackwell, J. & Ha ie, J. Feedback in music pe o mance
eaching. F on ie s in Psychology 13, 891025 (2022). URL h ps://www.
on ie sin.o g/a icles/10.3389/ psyg.2022.891025/ ull.
[131] Wisniewski, B., Zie e , K. & Ha ie, J. The powe o eedback e isi ed: A
me a-analysis o educa ional eedback esea ch. F on ie s in psychology 10,
487662 (2020).
[132] Seu e , T., Hamm, V., Vog , A. & Rieme , V. The in e play o cogni i e load,
lea ne s’ esou ces and sel - egula ion. Educa ional Psychology Re iew 36, 50
(2024). URL h ps://link.sp inge .com/10.1007/s10648-024-09890-1.
[133] Ng, K. C. e al. 3d augmen ed mi o : a mul imodal in e ace o s ing
ins umen lea ning and eaching wi h ges u e suppo . In P oceedings o he
9 h in e na ional con e ence on Mul imodal in e aces, 339–345 (2007).
[134] Blanco, A. D., Tassani, S. & Rami ez, R. Real- ime sound and mo ion eedback
o iolin bow echnique lea ning: A con olled, andomized ial. F on ie s
in Psychology 12, 648479 (2021).
[135] P o enzale, C., Di Tommaso, F., Di S e ano, N., Fo mica, D. & Ta oni, F.
Real- ime isual eedback based on mimus echnology educes bowing e o s
in beginne iolin s uden s. Senso s 24, 3961 (2024). URL h ps://www.
mdpi.com/1424-8220/24/12/3961.
[136] Ço lu, M., Mulle , C., Desme , F. & Leman, M. The consequences o
addi ional cogni i e load on pe o ming musicians. Psychology o Music
BIBLIOGRAPHY 105
43, 495–510 (2015). URL h ps://jou nals.sagepub.com/doi/10.1177/
0305735613519841.
[137] Michałko, A., Campo, A., Nijs, L., Leman, M. & Van Dyck, E. Towa d
a meaning ul echnology o ins umen al music educa ion: Teache s’ oice.
F on ie s in Educa ion 7, 1027042 (2022). URL h ps://www. on ie sin.
o g/a icles/10.3389/ educ.2022.1027042/ ull.
[138] Rami ez-Melendez, R. & Waddell, G. Technology-Enhanced Lea ning o
Pe o mance. In McPhe son, G. E. (ed.) The Ox o d Handbook o Music
Pe o mance, Volume 2, 0 (Ox o d Uni e si y P ess, 2022). URL h ps:
//doi.o g/10.1093/ox o dhb/9780190058869.013.24.
[139] Meissne , H. & Timme s, R. Teaching young musicians exp essi e pe o -
mance: an expe imen al s udy. Music Educa ion Resea ch 21, 20–39 (2019).
URL h ps://www. and online.com/doi/ ull/10.1080/14613808.2018.
1465031.
[140] Linds öm, E., Juslin, P. N., B esin, R. & Williamon, A. “exp essi i y comes
om wi hin you soul”: A ques ionnai e s udy o music s uden s’ pe spec i es
on exp essi i y. Resea ch S udies in Music Educa ion 20, 23–47 (2003). URL
h ps://jou nals.sagepub.com/doi/10.1177/1321103X030200010201.
[141] Sloboda, J. Explo ing he musical mind: Cogni ion, emo ion, abili y, unc ion
(Ox o d Uni e si y P ess, 2005).
[142] Meissne , H. Ins umen al eache s’ ins uc ional s a egies o acili a ing
child en’s lea ning o exp essi e music pe o mance: An explo a o y s udy.
In e na ional Jou nal o Music Educa ion 35, 118–135 (2017). URL h ps:
//jou nals.sagepub.com/doi/10.1177/0255761416643850.
[143] Has on, W. Beginning wind ins umen ins uc ion: A compa ison o au al
and isual app oaches. Con ibu ions o Music Educa ion 37, 9–28 (2010).
URL h p://www.js o .o g/s able/24127224.
106 BIBLIOGRAPHY
[144] Wul , G., McNe in, N. & Shea, C. H. The au oma ici y o complex mo o
skill lea ning as a unc ion o a en ional ocus. The Qua e ly Jou nal o
Expe imen al Psychology Sec ion A 54, 1143–1154 (2001). URL h ps://
doi.o g/10.1080/713756012. Publishe : SAGE Publica ions.
[145] Ebie, B. D. The e ec s o e bal, ocally modeled, kines he ic, and audio-
isual ea men condi ions on male and emale middle-school ocal mu-
sic s uden s’ abili ies o exp essi ely sing melodies. Psychology o Music
32, 405–417 (2004). URL h ps://jou nals.sagepub.com/doi/10.1177/
0305735604046098.
[146] Lisboa, T. Ac ion and hough in cello playing: An in es iga ion o child en’s
p ac ice and pe o mance. In e na ional jou nal o music educa ion 26, 243–
267 (2008).
[147] Duke, R. A., Cash, C. D. & Allen, S. E. Focus o A en ion A ec s Pe o mance
o Mo o Skills in Music. Jou nal o Resea ch in Music Educa ion 59, 44–
55 (2011). URL h ps://doi.o g/10.1177/0022429410396093. Publishe :
SAGE Publica ions Inc.
[148] Wul , G. A en ional ocus and mo o lea ning: a e iew o 15 yea s. In-
e na ional Re iew o Spo and Exe cise Psychology 6, 77–104 (2013). URL
h ps://doi.o g/10.1080/1750984X.2012.723728. Publishe : Rou ledge.
[149] Meissne , H. Theo e ical amewo k o acili a ing young musicians’
lea ning o exp essi e pe o mance. F on ie s in Psychology 11, 584171
(2021). URL h ps://www. on ie sin.o g/a icles/10.3389/ psyg.
2020.584171/ ull.
[150] B une , J. S., Ol e , R. R., G een ield, P. M. e al. S udies in cogni i e g ow h.
(1966).
[151] Hallam, S. & Papageo gi, I. Concep ions o musical unde s anding. Resea ch
S udies in Music Educa ion 38, 133–154 (2016).
BIBLIOGRAPHY 107
[152] Woody, R. H. Explaining exp essi e pe o mance: Componen cogni i e skills
in an au al modeling ask. Jou nal o Resea ch in Music Educa ion 51, 51–63
(2003). URL h ps://jou nals.sagepub.com/doi/10.2307/3345648.
[153] McPhe son, G. E., Osbo ne, M. S., E ans, P. & Miksza, P. Applying sel -
egula ed lea ning mic oanalysis o s udy musicians’ p ac ice. Psychology
o Music 47, 18–32 (2019). URL h ps://jou nals.sagepub.com/doi/10.
1177/0305735617731614.
[154] Waddell, G., Pe kins, R. & Williamon, A. The E alua ion Simula o : A New
App oach o T aining Music Pe o mance Assessmen . F on ie s in Psychology
10 (2019). URL h ps://www. on ie sin.o g/jou nals/psychology/
a icles/10.3389/ psyg.2019.00557.
[155] Shulman, L. S. Those who unde s and: Knowledge g ow h in eaching. Edu-
ca ional Resea che 15, 4–14 (1986). URL h ps://jou nals.sagepub.com/
doi/10.3102/0013189X015002004.
[156] Mac ides, E. & Angeli, C. Domain-speci ic aspec s o echnological peda-
gogical con en knowledge: Music educa ion and he impo ance o a ec .
TechT ends 62, 166–175 (2018). URL h p://link.sp inge .com/10.1007/
s11528-017-0244-7.
[157] Mac ides, E. & Angeli, C. In es iga ing pck h ough music ocusing on
a ec . The In e na ional Jou nal o In o ma ion and Lea ning Technology
35, 181–198 (2018). URL h p://www.eme ald.com/ijil /a icle/35/3/
181-198/136592.
[158] Juslin, P. N. & Sloboda, J. In oduc ion: aims, o ganiza ion, and e minology.
In Juslin, P. N. & Sloboda, J. (eds.) Handbook o Music and Emo ion: Theo y,
Resea ch, Applica ions, 3–12 (Ox o d Uni e si y P ess, Ox o d, 2011).
[159] Bu na d, P. & Younke , B. A. P oblem-sol ing and c ea i i y: Insigh s om
s uden s’ indi idual composing pa hways 22, 59–76 (2004). URL h ps://
jou nals.sagepub.com/doi/10.1177/0255761404042375.
108 BIBLIOGRAPHY
[160] Coulson, A. N. & Bu ke, B. M. C ea i i y in he elemen a y music class-
oom: A s udy o s uden s’ pe cep ions 31, 428–441 (2013). URL h ps:
//jou nals.sagepub.com/doi/10.1177/0255761413495760.
[161] Todd, J. R. & Mish a, J. Making lis ening ins uc ion meaning ul: A li e a u e
e iew. Upda e: Applica ions o Resea ch in Music Educa ion 31, 4–10 (2013).
URL h ps://jou nals.sagepub.com/doi/10.1177/8755123312473609.
[162] Swanwick, K. & Ca alie i F anca, C. Composing, pe o ming and audience-
lis ening as indica o s o musical unde s anding. B i ish Jou nal o Music Ed-
uca ion 16, 5–19 (1999). URL h ps://www.camb idge.o g/co e/p oduc /
iden i ie /S026505179900011X/ ype/jou nal_a icle.
[163] Mac ides, E. & Angeli, C. Music cogni ion and a ec in he design
o echnology-enhanced music lessons. F on ie s in Educa ion 5, 518209
(2020). URL h ps://www. on ie sin.o g/a icle/10.3389/ educ.
2020.518209/ ull.
[164] Gab ielsson, A. The Rela ionship be ween Musical S uc u e and Pe cei ed
Exp ession (Ox o d Uni e si y P ess, 2016). URL h ps://academic.oup.
com/edi ed- olume/34489/chap e /292614599.
[165] Pew Resea ch Cen e . Ame icans’ use o mobile echnology and home b oad-
band (2024). URL h ps://pew s .ch/49i9IFc.
[166] Eu opean Commission. E-communica ions su ey (2021). URL h ps:
//eu opa.eu/eu oba ome e /su eys/de ail/2232. Fieldwo k conduc ed
No embe 2020 - Decembe 2020. Reques ed by Communica ions Ne wo ks,
Con en and Technology. Re e ence: SP510.
[167] Tho ge sen, K. & Zandén, O. The in e ne as eache . Jou nal o
Music, Technology amp; Educa ion 7, 233–244 (2014). URL h ps://
in ellec disco e .com/con en /jou nals/10.1386/jm e.7.2.233_1.
[168] Chisholm, M. A. e al. In luence o in e ac i e ideocon e encing on he pe o -
mance o pha macy s uden s and ins uc o s. Ame ican Jou nal o Pha ma-
BIBLIOGRAPHY 109
ceu ical Educa ion 64, 152 (2000). URL h ps://www.lea n echlib.o g/
p/90385.
[169] Nguyen, T. The e ec i eness o online lea ning: Beyond no signi ican di e -
ence and u u e ho izons. MERLOT Jou nal o online lea ning and eaching
11, 309–319 (2015).
[170] Cao, Y. Online cou ses o piano eaching: compa ing he e ec i eness
and impac o mode n and adi ional me hods. Educa ion and In o ma-
ion Technologies 29, 11407–11419 (2024). URL h ps://doi.o g/10.1007/
s10639-023-12283-6.
[171] Chiu, T. K. & Hew, T. K. Fac o s in luencing pee lea ning and pe o mance in
MOOC asynch onous online discussion o um. Aus alasian Jou nal o Educa-
ional Technology 34 (2018). URL h ps://aje .o g.au/index.php/AJET/
a icle/ iew/3240. Sec ion: A icles.
[172] Chiu, T. K. F., Lin, T.-J. & Lonka, K. Mo i a ing Online Lea ning: The
Challenges o COVID-19 and Beyond. The Asia-Paci ic Educa ion Resea che
30, 187–190 (2021). URL h ps://doi.o g/10.1007/s40299-021-00566-w.
[173] S ickland, S., Sco , N. & A hauda, R. A amewo k o eal- ime online col-
labo a ion in music p oduc ion. In P oceedings o he ACMC2018: Con e ence
o he Aus alasian Compu e Music Associa ion, 79–86 (2018).
[174] Volio i, G. & Williamon, A. Reco dings as lea ning and p ac ising e-
sou ces o pe o mance: Explo ing a i udes and beha iou s o music s u-
den s and p o essionals. Musicae Scien iae 21, 499–523 (2017). URL h ps:
//jou nals.sagepub.com/doi/10.1177/1029864916674048.
[175] Hamond, L. F., Welch, G. & Himonides, E. The Pedagogical Use o Visual
Feedback o Enhancing Dynamics in Highe Educa ion Piano Lea ning and
Pe o mance. OPUS; . 25, n. 3 (2019): Volume 25, núme o 3. Se emb o-
dezemb o 2019.DO - 10.20504/opus2019c2526 (2019). URL h ps://www.
anppom.com.b / e is a/index.php/opus/a icle/ iew/opus2019c2526.
110 BIBLIOGRAPHY
[176] Exp essi eness in music pe o mance: Empi ical app oaches ac oss s yles
and cul u es (Ox o d Uni e si y P essOx o d, 2014), 1 edn. URL h ps:
//academic.oup.com/book/27746.
[177] Tejada, J. & Fe nández-Villa , M. Design and Valida ion o So wa e o he
T aining and Au oma ic E alua ion o Music In ona ion on Non-Fixed Pi ch
Ins umen s o No ice S uden s. Educa ion Sciences 13 (2023).
[178] Tejada, J., Mu illo Ribes, A., Mo an Na asquillo, R., Be nabé Villod e, M.
d. M. & Fe nández-Vila , M. Music eache s’ pe cep ions o a new so wa e
o he assessmen o musical ins umen in ona ion (2024). URL h ps:
//hdl.handle.ne /10550/108659.
[179] Rami ez, R. e al. Enhancing music lea ning wi h sma echnologies. In
P oceedings o he 5 h In e na ional Con e ence on Mo emen and Compu -
ing, MOCO ’18 (Associa ion o Compu ing Machine y, New Yo k, NY, USA,
2018). URL h ps://doi.o g/10.1145/3212721.3212886.
[180] Upham, F. & McAdams, S. Ac i i y analysis and coo dina ion
in con inuous esponses o music. Music Pe cep ion 35, 253–294
(2018). URL h ps://online.ucp ess.edu/mp/a icle/35/3/253/91976/
Ac i i y-Analysis-and-Coo dina ion-in-Con inuous.
[181] Sun, L., Hu, L., Ren, G. & Yang, Y. Musical ension associa ed wi h
iola ions o hie a chical s uc u e. F on ie s in Human Neu oscience 14,
578112 (2020). URL h ps://www. on ie sin.o g/a icle/10.3389/
nhum.2020.578112/ ull.
[182] Schube , E. & Wol e, J. Does imb al b igh ness scale wi h equency and
spec al cen oid. Ac a Acus ica Uni ed Wi h Acus ica 92, 820–825 (2006).
URL h ps://api.seman icschola .o g/Co pusID:8061120.
[183] Mela a, R. D. & Ma ks, L. E. In e ac ion among audi o y dimensions: Timb e,
pi ch, and loudness. Pe cep ion & Psychophysics 48, 169–178 (1990). Place:
US Publishe : Psychonomic Socie y.
BIBLIOGRAPHY 111
[184] Williamon, A., Waddell, G. & Moghnieh, A. D2.5 – use e alua ions o each
echnology in each scena io. Tech. Rep. TELMI-D-WP2-RCM-20181031-D2.5,
Royal College o Music, TELMI P ojec (2019). URL h ps:// elmi.up .
edu/. Ho izon 2020 Resea ch and Inno a ion Ac ion, G an Ag eemen No.
688269.
[185] Ra el, C. & Ellis, D. P. W. In ui i e analysis, c ea ion and manipula ion o
midi da a wi h p e y_midi. In P oceedings o he 15 h In e na ional Con e -
ence on Music In o ma ion Re ie al La e B eaking and Demo Pape s (2014).
[186] Bjø ndalen, O. M. & Dou senaud, R. Mido: MIDI Objec s o Py hon. h ps:
//mido. ead hedocs.io (2023). URL h ps://mido. ead hedocs.io. Ve -
sion 1.3.2.
[187] Bianchini, S., Hanappe, P. & Lee, J. FluidSyn h: Real-Time So wa e
Syn hesize . h ps://www. luidsyn h.o g/ (2025). URL h ps://www.
luidsyn h.o g/. C oss-pla o m so wa e syn hesize based on he Sound-
Fon 2 speci ica ion.
[188] McFee, B. e al. lib osa: Audio and Music Signal Analysis in Py hon. h ps:
//lib osa.o g/ (2023). Ve sion 0.10.2.pos 1.
[189] Kim, J. W., Salamon, J., Li, P. & Bello, J. P. C epe: A con olu ional ep-
esen a ion o pi ch es ima ion. In P oceedings o he IEEE In e na ional
Con e ence on Acous ics, Speech, and Signal P ocessing (ICASSP), 161–165
(2018).
[190] Zapa a, J., Holzap el, A., Da ies, M., Oli ei a, J. & Gouyon, F. Assigning
a con idence h eshold on au oma ic bea anno a ion in la ge da ase s. In
In e na ional Socie y o Music In o ma ion Re ie al Con e ence (ISMIR),
157–162 (2012).
[191] Zapa a, J., Da ies, M. & Gómez, E. Mul i- ea u e bea acke . IEEE/ACM
T ansac ions on Audio, Speech, and Language P ocessing 22, 816–825 (2014).
118 Appendix A. Appendix: MuSA E alua ion S udy Resul s
Table 18: Rep esen a i e pa icipan quo es by ques ion and sen imen .
Ques ion Sen imen Quo e
Expe ience Posi i e (n7oh8onc) I was good! I de ini ely was be e han
jus plain p ac ice. I liked he highligh ed egions
whe e i de ec ed po en ial inconsis encies.
Expe ience Posi i e (nd qs3sg) I hink ha he eedback is use ul o en-
gagemen and inc ease he desi e o imp o emen . I
could be use ul, a some poin ha you ha e he g aph-
ics and y o adjus o hem in eal ime.
Expe ience Nega i e (chx j0wq) The eedback ool was no in eal- ime so
I could no make immedia e co ec ions. I was no
e y easy o adjus my pi ch a e he analysis ool.
The eedback ool was use ul in unde s anding whe e
I was co ec and whe. . .
Expe ience Nega i e (y6 3g63s) The ool looks g ea bu seems o ha e
some p oblem wi h alignmen s be ween wha I sang
and wha was supposed o be sang.
Imp o emen Posi i e (y6 3g63s) Fo su e!
Imp o emen Posi i e (b068ba w) Yes, I ocused mo e on nuances, he ha d-
es one was empo because I didn’ know i I was ush-
ing o no .
Imp o emen Nega i e (opl 7i5e) No eally because o me audi o y cues a e
mo e help ul han isual cues.
Imp o emen Nega i e (3d1ly0bg) I ’s ha d o say, I hope he e’s a con ol
g oup ha is jus p ac icing and lis ening o hei own
audio, because I eel like he p ac ice inhe en ly helps.
Highligh s Posi i e (w4m19emj) Yes hey we e help ul o showing me
whe e I need o ocus.
Highligh s Posi i e (xzdbn2c2) Yes de ini ely. The pi ch cu e and dy-
namics cu e we e qui e good choices o he plo .
Highligh s Nega i e (3 zis ce) They we e, bu some imes I wouldn’ see
any highligh ed a ea e en hough I s ill conside ed he
pe o mance no pe ec a all.
Highligh s Nega i e (3d1ly0bg) Yeah hey we e.
Paymen Posi i e (e uygx3k) I I we e eco ding music I would de ini ely
pay o i because I hink i is e y use ul o imp o ing
bu a he momen I don’ ha e a need o i .
Paymen Posi i e (nd qs3sg) I my wo k implied singing, yes, bu I would
compa e i i s o o he simila ools i exis s. I
singing is a hobby I would no pay o i .
Paymen Nega i e (w4m19emj) I would pay a maximum o 5$ a mon h.
Paymen Nega i e (opl 7i5e) No.
119
Table 19: Summa y o sen imen analysis esul s, including o e all s a is ics, pe -
ques ion means and medians, and co ela ion be ween lexicon and VADER sen i-
men .
Ca ego y Sen imen Mean Median N / No es
O e all Sen imen S a is ics
O e all Lexicon 0.223 0.200 56
O e all VADER 0.370 0.458 56
Sen imen by Ques ion
Expe ience Lexicon 0.308 0.306 14
Expe ience VADER 0.671 0.795 14
Imp o emen Lexicon 0.186 0.191 14
Imp o emen VADER 0.286 0.402 14
Highligh s Lexicon 0.242 0.200 14
Highligh s VADER 0.282 0.421 14
Paymen Lexicon 0.156 0.000 14
Paymen VADER 0.241 0.095 14
Co ela ion
Lexicon s VADER Pea son 0.549 — —
