WHAT SONG NOW?
PERSONALIZED RHYTHM GUITAR LEARNING
IN WESTERN POPULAR MUSIC
Zaka ia Hassein-Bey1Yohann Abbou2Alexand e D’Hooge1
Ma hieu Gi aud1Gilles Guillemain2Au élien Jeanneau1
1Uni . Lille, CNRS, Cen ale Lille, UMR 9189 CRIS AL, F-59000 Lille, F ance
2Gui a Social Club, F-59000 Lille, F ance
ABSTRACT
The gui a is one o he mos popula musical ins umen s,
and nume ous pedagogical ools ha e been de eloped o
suppo lea ne s. They ely on as collec ions o songs,
shee music, and abla u es, making i challenging o gui-
a is s o na iga e and iden i y pieces ha a e bo h peda-
gogically ele an and aligned wi h hei musical in e es s.
We in oduce a simple mul i-c i e ia ule-based model
o assess bo h he di icul y o lea ning a piece and he
skill le el o a gui a is , aking in o accoun musical and
echnical c i e ia. The model p o ides pe sonalized sug-
ges ions ha help lea ne s p og ess e icien ly, conside -
ing pa s wi hin songs, bu also mul iple e sions o he
same pa , accoun ing o simpli ied adap a ions o di e -
en playing s yles, and inally exe cises used o p og es-
si ely lea n each pa e sion.
We implemen and e alua e his app oach in he con ex
o hy hm gui a in popula music, using a da ase designed
o he p op ie a y applica ion Gui a Social Club. Expe
e alua ion o 77 ecommenda ions o eigh use p o iles o
a ying le els indica e ha in 82% o cases, he model p o-
ides ele an sugges ions. While he ull da ase emains
p op ie a y, we elease unde open licenses he code along
wi h a sub-co pus con aining anno a ed di icul ies o 319
e sions o 110 pa s om 40 songs.
1. INTRODUCTION
Gui a pedagogy has e ol ed alongside he echnologies
a ailable o he gene al public [1]. Video-sha ing pla -
o ms ha e enabled he o ma ion o digi al and eal-li e
lea ning communi ies [2]. Al hough he ac ual pedagogi-
cal alue o “se ious games” is some imes deba ed [3, 4],
games such as Gui a He o,RockBand, o Rocksmi h ha e
con ibu ed o popula izing he gui a among he gene al
public. Such games can in ol e con olle s, eal gui a s,
and/o augmen ed eali y [5,6].
© Z. Hassein-Bey, Y. Abbou, A. D’Hooge, M. Gi aud, G.
Guillemain, and A. Jeanneau. Licensed unde a C ea i e Commons A i-
bu ion 4.0 In e na ional License (CC BY 4.0). A ibu ion: Z. Hassein-
Bey, Y. Abbou, A. D’Hooge, M. Gi aud, G. Guillemain, and A. Jeanneau,
“Wha Song Now? Pe sonalized Rhy hm Gui a Lea ning in Wes e n
Popula Music”, in P oc. o he 26 h In . Socie y o Music In o ma ion
Re ie al Con ., Daejeon, Sou h Ko ea, 2025.
The as numbe o ools and esou ces a ailable o mu-
sicians makes selec ing sui able pedagogical ma e ial com-
plex. Solu ions ha e been p oposed o compile lea ning e-
sou ces ela ed o popula songs [7] o o sea ch o pieces
based on he cho ds use s wish o p ac ice [8]. Song ec-
ommenda ion can be op imized o maximize he numbe o
playable songs a e lea ning a new cho d [9]. One o he
majo challenges in hese applica ions is sugges ing pieces
o use s ha ma ch hei skill le el, which includes chal-
lenges in bo h es ima ing he song di icul y and he use
le el.
Models wi h mul iple c i e ia o di icul y es ima ion
ha e been p oposed o piano music [10]. Recen wo k
also p oposed e alua ing he di icul y o a piece by es i-
ma ing a inge ing o di ec ly analyzing an image o he
sco e [11, 12]. Fo he gui a , Vasquez e al. in e iewed
gui a eache s o de elop c i e ia o es ima ing he di i-
cul y o a piece wi h se en 4-scale c i e ia, along wi h a
machine lea ning me hod o au oma e human a ings [13].
Con ibu ions. Pedagogical assis ance is c ucial in
lea ning music. Howe e , o ou knowledge, no app oach
cu en ly p o ides an op imized lea ning pa h ha balances
di icul y and in e es o gui a is s o all le els. Ou goal
is o sugges songs ailo ed o lea ne s’ cu en le el and
pas expe ience, ensu ing bo h accessibili y and p og ess.
Focusing on accompanimen o hy hm gui a [14], we
iden i y wha makes he di icul y o a gui a is , based on
mul iple c i e ia, and conside ing songs, pa s, and e -
sions (Sec ion 2). We p esen a da ase o 1344 e sions o
534 song pa s (Sec ion 3), and p opose a model o sugges
e sions, pa s, and songs pe sonalized o someone’s skills
(Sec ion 4). This model is in eg a ed in o he p op ie a y
applica ion Gui a Social Club (GSC), which is cu en ly
in de elopmen , aimed o c ea e an en i onmen cen e ed
a ound gui a playing and lea ning. We implemen and es
his model (Sec ion 5), and conclude wi h a discussion on
pe sonalized pedagogy (Sec ion 6).
2. WHAT MAKES LEARNING NEW SONGS
DIFFICULT?
2.1 Di icul y C i e ia and Exe cises
The skill le el o a lea ne canno be de ined wi h a single
alue. A gui a is may, o ins ance, know many cho ds,
296
`CCho d complexi y Cho d examples
0-2 Basic open cho ds, simple oicings, ew inge s wi h close posi ions Em, E, Am, A7sus4
3-4 Open cho ds, ligh in e sions, mo e open posi ions C, D7, Gsus4, Fadd9
5-6 Common ba e cho ds, mode a e s e ches F, Dmin7, G9, Bbadd6
7-8 Mo e ba e posi ions, less in ui i e g ips EM9, F#m, C#m/G#, Eb
9-10 Ad anced oicings, la ge s e ches, ull e boa d usage Fm7, Abm, C#/G, Esus4/B
`PRhy hmic Pa e ns
0-2 Mos ly hal and qua e no es
3-4 Mos ly eigh s, some six een hs
5-6 Six een hs wi h egula pa e ns
7-8 Six een hs, i egula pa e ns, cla es, poly hy hms
9-10 Elabo a ed poly hy hms, odd- ime signa u es
`TTempo / Impac Complexi y
0-2 Slow (60-200 PIPM)
3-4 Medium (200-400 PIPM)
5-6 Fas e (400-600 PIPM)
7-8 High-speed (600-800 PIPM)
9-10 Ex eme (>800 PIPM)
`GTypical ime Global Di icul y
0-2 T0 Absolu e beginne : basic open cho ds, simple s umming, slow empo
3-4 T0 + 6 mon hs Beginne : basic open cho ds, simple s umming, slow empo
5-6 T0 + 3 yea s In e media e: ba e cho ds, basic hy hm a ia ions, mode a e empo
7-8 T0 + 5 yea s Con i med: complex cho d ansi ions, in ica e hy hms, as e playabili y
9-10 T0 + 8 yea s Ad anced: mas e y o mul iple s yles, imp o isa ion, echnical p o iciency
Expe : nea -p o essional le el, ull a is ic con ol, ad anced imp o isa ion
Table 1. C i e ia o assessing song/pa / e sion di icul y in hy hm gui a playing, as discussed in Sec ion 2. (a) Cho ds.
(b) Rhy hmic Pa e ns. The e a e u he co ec ing ac o s o bina y/ e na y me e s, upbea s... No e ha his co pus mos ly
con ain pop songs wi hou elabo a ed poly hy hms. (c) Peak Impac s Pe Minu e (PIPM). No e ha e y o ex emely slow
PIPM can be also ha d o play (no displayed he e). (d) Global di icul y. Lea ning imes a y conside ably among s uden s,
and some ne e p og ess beyond he Beginne o In e media e le els. The du a ions men ioned ypically apply o s uden s
who p ac ice hei ins umen se e al imes a week.
including complex ones, bu s uggle wi h smoo h ansi-
ions. Con e sely, ano he gui a is may only know basic
cho ds bu swi ch be ween hem e o lessly, e en in com-
plex hy hms. The au ho s o [13] iden i y se e al di i-
cul y c i e ia: he in insic complexi y o a cho d based
on i s inge ing, he a i y o a cho d, how he cho d is
s ummed, and wi hin a cho d sequence, he epe i i eness
and speed o cho d ansi ions.
The c i e ia p oposed in his pape (Table 1) a e also
based on he expe ise o a gui a eache (YA, second au-
ho ). While many c i e ia a e simila , some di e ences
a ose, especially when i comes o he ela i e impo ance
o each aspec . I ’s wo h no ing ha ou wo k is ocused
abou assis ing gui a lea ning while [13] aimed a p e-
dic ing he playabili y o gui a songs in a mo e gene al
ashion, which migh explain a ia ions in he c i e ia.
Cho d complexi y (`C) can be es ima ed using se e al
indica o s: he numbe o inge s equi ed, he e span,
he use o open o mu ed s ings, and he o e all hand po-
si ioning. Howe e , he ue challenge o en lies no in
playing a single cho d, bu in ansi ioning be ween cho ds
and hand posi ions. A key pedagogical p inciple empha-
sized by he gui a eache is ha o a leading inge . An
open C majo cho d ypically uses he ing inge as an an-
cho , while D/F# o en equi es he index inge o e en he
humb. Al e na ing be ween cho ds wi h di e en leading
inge s – such as in “C o D/F#” o e en in he e y com-
mon “G o C” – demands mo e e o and p ecision om
he lea ne . In con as , ansi ions like “Em o Asus2” a e
ela i ely easy, as hey in ol e minimal inge mo emen
and main ain a consis en hand shape (Figu e 1).
Conside ing empo (`T), he amoun o Bea s Pe
Minu e (BPM, conside ing qua e no es as bea s) alone is
X O O
C
3 2 1
X X
D/F#
1 2 4 3
O O
G
2 1 3 4
X O O
C
3 2 1
O O O O
Em
2 3
X O O O
Asus2
2 3
Figu e 1. Gui a cho d diag ams wi h possible inge ings
(1: index inge ) and ansi ions. C o D/F# equi es a com-
ple e inge eposi ioning. G o C in ol es mo ing he in-
dex and middle inge s oge he . Em o Asus2 is easie ,
equi ing only a shi o he hand owa d highe s ings.
no su icien o cha ac e ize empo complexi y. Fo in-
s ance, a song a 70 BPM whe e he gui a is p ima ily
plays six een h no es can p esen a simila le el o hy h-
mic challenge as a song a 140 BPM wi h mos ly eigh h
no es. In bo h cases, he e ec i e playing a e is compa a-
ble, e en i he pe cei ed speed is di e en o he playe
and he lis ene . To be e cap u e his aspec and o cha ac-
e ize he y hmic densi y o a piece, we in oduce he Peak
Impac s Pe Minu e (PIPM) – ha is, he maximum bea -
a e (pe minu e) o he mos equen ly occu ing hy hmic
igu e. Bo h p e ious examples would be a 280 PIPM,
e en i hey would ea u e pa s wi h lowe PIPM.
Rhy hmic pa e ns (`P) also play a signi ican ole in de-
e mining di icul y – pa icula ly hei egula i y and a -
i y. Su p isingly, hy hmic pa e ns played almos no ole
in [13] o di icul y es ima ion.
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
297
1. Knocking on Hea en’s Doo (Bob Dylan, 1973)
Ve sion a Ve sion b Ve sion c Ve sion d (SR) Ve sion e (OR)
`C`P`T`G`C`P`T`G`C`P`T`G`C`P`T`G`C`P`T`G
In o/Cho us 1.a o/ 120 1 0 0 0 1.b
,
/ 90 1 1 1 0 1.c
C
/ 120 1 1 1 1 1.d
/ 72 1 2 2 1 1.e
/ 72 1 2 2 1
Ve se 2.a o/ 120 3 0 0 1 2.b
,
/ 120 3 1 1 2 2.c
C
/ 120 3 1 1 2 2.d
/ 72 3 2 2 2 2.e
/ 72 3 2 2 3
Full Song 0.a o/ 72 3 0 0 1 0.b
,
/ 72 3 1 1 2 0.c
C
/ 72 3 1 1 2 0.d
/ 72 3 2 2 2 0.e
/ 72 3 2 2 3
65. Losing my Religion (R.E.M., 1991)
Ve sion c Ve sion d (SR) Ve sion e (OR)
`C`P`T`G`C`P`T`G`C`P`T`G
In o/Cho us 1.c
C
/ 120 4 1 1 3 1.d
/ 90 4 2 3 3 1.e
/ 125 4 2 4 4
Ve se 2.c
C
/ 120 2 1 1 2 2.d
/ 90 2 2 3 2 2.e
/ 125 4 2 4 3
B idge 4.c
C
/ 120 4 1 1 2 4.d
/ 90 4 2 1 2 4.e
/ 125 4 2 1 2
Full Song 0.c
C
/ 125 4 1 1 3 0.d
/ 125 4 2 3 3 0.e
/ 125 4 2 4 4
55. Fade To Black (Me allica, 1984)
Ve sion e (OR)
`C`P`T`G
In o 1.e
/ 58 2 3 3 4
B idge 0 2.e
/ 90 9 3 3 7
Ve se 3.e
C
/ 120 3 3 3 4
Cho us 4.e
©
/ 58 3 3 6 5
B idge 1 5.e
©
/ 58 3 3 7 6
B idge 2 6.e
©
/ 58 3 3 7 5
Ou o 7.e Cla e / 58 4 4 5 5
Full Song 0.e OR / 58 9 4 7 8
Table 2. Ra ings o selec ed e sions and pa s om h ee songs. Ve sions (a), (b), and (c) a e in oduc o y a angemen s,
using mainly whole, hal , and qua e -no e hy hmic pa e ns, espec i ely. Ve sion (d) is close o he o iginal song bu
ea u es a Simpli ied Rhy hm (SR), educing he complexi y o he mos challenging pa e ns. Ve sion (e) p esen s he
O iginal Rhy hm (OR), closely ep oducing he e sion pe o med by he o iginal band.
C
/120 indica es ha he e sion
has a as es no e alue o a qua e no e and played wi h a empo o 120 BPM. Knocking on Hea en’s Doo can be played
by beginne s, including by absolu e beginne s o i s e sion (a), whe e cho ds change on each whole no e (`G= 1). In
Losing my eligion, in easy e sions, In o and Cho us a e mo e challenging han he Ve se, especially conce ning cho ds
(`C). Fade o Black, wi h i s 7 pa s, is des ina ed o ad anced playe s. The “B idge 0”, ea u ing diminished cho ds wi h
wo di e en oicings, is pa icula ly di icul (`C= 9). As his song ea u es e y cha ac e is ic oicings and elabo a e
ansi ions, he decision was o ha e only he e sion (e) in he co pus.
Di icul y also a ises om he in e ac ion be ween he
le and igh hands, pa icula ly when he numbe o
s ings plucked o s ummed a ies signi ican ly be ween
cho ds. This combina o ial complexi y u he con ibu es
o he challenge o cho d ansi ions. A model o hy hmic
gui a di icul y should conside no only he complexi y
o indi idual cho d posi ions o hy hmic pa e ns, bu also
he di icul y o ansi ions be ween hem, especially a a
gi en empo. Al oge he , each song di icul y is a ed as a
whole (`G) as pe cei ed by he gui a eache .
Mo eo e , when wo king wi h a eache , s uden s a e
o en gi en speci ic exe cises o a ge challenging aspec s
o a song – mos commonly pa icula cho d o pa e n
changes ha need o be p ac iced a a ying speeds. In his
way, di e en songs may sha e common exe cises. As a e-
sul , he di icul y o lea ning a new song also depends on
he numbe and complexi y o new exe cises i in oduces,
beyond hose he s uden has al eady mas e ed.
2.2 Songs, Pa s, and Ve sions
E en wi h mul iple c i e ia, ea ing he di icul y o a song
as ixed is un ealis ic, since i a ies h oughou he piece.
R.E.M.’s Losing my Religion, o example, has a B idge
ha is mo e echnical and challenging han he Ve se be-
cause o he as -paced cho d changes (Table 2). Indeed,
he Ve se is mos ly composed o he simple sequence o
cho ds (Am, Em) which is a pe ec o a e y i s piece o
a gui a beginne . The B idge howe e con ains he cho ds
G, F, Dm, and Am, and hus ansi ions be ween hem. The
F cho d is pa icula ly complex o beginne s because i
has o be played wi h a ba e.
Bu wha does i eally mean o “play Losing My Re-
ligion”? Some ad anced playe s may aim o mas e he
ull abla u e as pe o med by he band, while o many
gui a is s, simply s umming he Ve se cho ds – enough o
sing along – is al eady a meaning ul goal. This has eal
pedagogical alue. A beginne who canno ye play he
Ve se as eco ded may s ill eel accomplished lea ning a
simpli ied e sion, guided by hei eache in choosing an
adap a ion sui ed o hei le el. Thus, a song and e en each
pa o a song could ha e mul iple e sions, each wi h an
es ima ed di icul y le el ha e lec s di e en app oaches
o playing i (Table 2). Simpli ica ion can ocus on educ-
ing he numbe o complexi y o cho ds – o example, by
emo ing ba e cho ds – as well as simpli ying s umming
pa e ns o o he musical ea u es.
We hus conside songs,pa s, e sions, and exe cises
linked o hem. The segmen a ion o a song in o pa s, such
as In o, Ve se, Cho us, B idge, o Ou o, is bo h musical
and echnical. Since pa s a e o en homogeneous om
a gui a is ic pe spec i e, i is possible o model hei own
di icul y and pedagogical in e es o a gi en pe son.
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
298
0 2 4 6 8 10
Cho ds
0
50
100
150
200
250
Ve sions
a
b
c
d
e
0 2 4 6 8 10
Rhy hmic Pa e ns
0
100
200
300
400
Ve sions
a
b
c
d
e
0 2 4 6 8 10
Tempo/impac
0
50
100
150
200
250
300
350
Ve sions
a
b
c
d
e
0 2 4 6 8 10
Global
0
50
100
150
200
250
Ve sions
a
b
c
d
e
Figu e 2. Dis ibu ion o he di icul y o he 1344 e sions among he ou c i e ia. Ve sion ange om (a), easies , o (e),
mos di icul , as desc ibed on Table 2.
3. CORPUS
The co pus de elopped o Gui a Social Club con ains
194 songs om he popula acous ic gui a epe oi e. The
gui a eache selec ed pieces equen ly eques ed by s u-
den s o a ious ages, gende s, and musical p e e ences.
The songs a e ca ego ized in o six gen es: Rock (32%),
Pop (29%), Funk/Soul/R&B (10%), Wo ld Music (9%),
and O he /Radio (20%). Expanding he co pus and be e
ep esen ing unde ep esen ed s yles is a u u e goal.
The songs con ain be ween 1 and 7 pa s, wi h mos
ha ing 2 o 3, o aling 534 pa s. The pa s ha e in a -
e age 2.5 e sions, o aling 1344 e sions. Finally, he
applica ion includes 3248 ideo exe cises. Mos o hese
exe cises ocus on p ac icing cho d ansi ions a a ying
speeds. Each exe cise was manually linked o speci ic e -
sions/pa s/songs o he da ase . The applica ion u he
con ains pseudo-songs wi h ins uc ional ideos on gene ic
opics.
4. A MODEL FOR PERSONALIZED
SUGGESTION
We conside a ini e co pus o songs S={s1, s2, ..., sS},
pa s P={p1, p2, ..., pP}, e sions V={ 1, 2, ..., V},
and exe cices E={e1, e2, ..., eE}, Each e sion is
linked o one pa π( )∈ P, and each pa pis linked
o one song σ(p)∈ S. Each e sion is also linked o a
se o exe cices ε( )⊂ E. A gui a is gal eady knows a
subse Vg⊂ V o hese e sions. We assume ha i means
ha hey al eady mas e he exe cises Eg=∪ ∈Vgε( ).
Which e sion o which pa o which piece, and hus
which exe cises, should hey wo k on o p og ess? To p o-
ide a pe sonalized pedagogical sugges ion, ou app oach
is o conside mul iple c i e ia o model he di icul y o
a e sion/pa (Sec ion 4.1) and o es ima e he gui a is ’s
skill le els (4.2). This allows us o model he pedagogical
alue o a e sion/pa o each lea ne (4.3).
4.1 Modeling Ve sion/Pa /Song Di icul y
Each song/pa / e sion was analyzed and assessed ac-
co ding o Kdi icul y c i e ia {`1, `2, . . . , `K}. Each c i-
e ion ep esen s a alue `k( )∈[0,10], whe e 0indica es
a e sion ha a beginne can play and 10 is o an expe-
ienced gui a is . The K= 4 e alua ed c i e ia desc ibed
in Sec ion 2 a e hus Cho ds (`C), Rhy hmic Pa e ns (`P),
Tempo/Impac (`T), and Global (`G) (Table 1).
C i e ion `Gwas assessed by he gui a eache , and
o he c i e ia combine human expe e alua ion ( ollowing
Table 1) wi h p op ie a y p ocedu es. Table 2 shows hese
a ings o h ee songs o inc easing di icul y, and Figu e 2
he dis ibu ion o a ings in he co pus.
Fo hy hm- ela ed c i e ia, mode a e and slow alues
p edomina e. The dis ibu ion o he di icul y c i e ia is
mos ly cen e ed a ound in e media e a ings. This obse -
a ion sugges s ha he co pus is well dis ibu ed ac oss
le els and could likely bene i gui a is s o all le els.
4.2 Es ima ing a Lea ne ’s Skill Le el
How can we desc ibe he skill le el o a gui a is gac oss
he conside ed c i e ia? Fo each c i e ion `k, we e ie e
Vk[N]
g⊂ Vg, he lis o he Nmos di icul known e sions
acco ding o ha c i e ion. The skill le el `k(g)o he gui-
a is is hen he a e age di icul y o hese N e sions:
`k(g) = 1
NX
∈Vk[N]
g
`k( )
Fo example, wi h N= 4, a gui a is gwho knows
pe ec ly Knocking on Hea en’s Doo (all pa s, e sion
(e)) bu only he easies e sion (c) o Losing my Religion
would ha e skill le els `C(g) = 3.75,`P(g) = 1.75,
`T(g) = 1.75, and `G(g) = 3.00. In he ac ual model,
he le els o each gui a is a e es ima ed by selec ing N=
15 pa e sions among he ones hey p ac iced, which
amoun s o 6 songs on a e age.
4.3 Pe sonalized Ve sion Sugges ions
The lea ne ’s skill le el is compa ed o he di icul y o
a speci ic e sion o a song pa , conside ing a challenge
alue τ∈[1,2,3,4] ha will be discussed below. A chal-
lenge i Fg( , τ)is compu ed o a gui a is gwishing o
lea n a e sion unde he challenge τ:
Fg( , τ) =
K•
X
k=1
αk|Ck(g, , τ)|
The challenge i is a weigh ed sum o he absolu e alue
o K• i s Ck(g, , τ), each weigh ed by a coe icien αk.
These i s ake in o accoun he Kdi icul y c i e ia de ined
p e iously (see Sec ion 4.3.1), bu also o he c i e ia (4.3.2,
4.3.3), and equal ze o when he i is pe ec .
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
299
1 0 1 2 3 4 5 6
Tempo/impac di icul y
0
10
20
30
40
50
Rank
1
2
3
4
1 0 1 2 3 4 5 6
Global di icul y
0
10
20
30
40
50
Rank
1
2
3
4
Figu e 3. Dis ibu ion o he ecommenda ions, on each o
he ou challenges acco ding o `Tand `G, o he p o ile
G4(wi h le els `T= 2.89 and `G= 3.33).
4.3.1 Challenge Fi o a Di icul y C i e ion
The i is modeled using an exponen ial unc ion wi h pa-
ame e s βk(scaling coe icien ) and γk,τ (bias o con ol
a ge di icul y unde challenge τ):
Ck(g, , τ) = eβk[`k( )−`k(g)−γk,τ ]−1
When γk,τ = 0, hen he alue Ck(g, , τ)equals
0when `k( ) = `k(g), meaning he e sion exac ly
ma ches he gui a is ’s le el. When γk,τ is s ic ly posi i e
( esp. nega i e), we wan ha he gui a is a ge a highe
( esp. lowe ) di icul y han hei cu en le el.
The model e en ually p oposes ou challenges (Ta-
ble 3). Challenge 2 sugges s pa s/ e sions a he gui a is ’s
le el (γk,2≈0), Challenge 1 sugges s easie pa s/ e sions
(γk,1nega i e), and Challenges 3 and 4 sugges p og es-
si ely ha de pa s/ e sions (γk,τ posi i e).
4.3.2 Challenge Fi o he New Exe cices
We call nεg( ) = |e∈ε( ) Eg| he numbe o exe cices
in ε( )no cu en ly known by g. Le γE,τ be he numbe
o new exe cices o be a ge ed, hen he i is:
CE(g, , τ) = eβE[nεg( )−γE,τ ]−1
This alue could be adap ed o accoun o he di icul y
o each exe cise, p o ided such da a is a ailable.
4.3.3 Addi ional Challenge Fi s
O he c i e ia kcan be implemen ed. Fo example, he p o-
p ie a y applica ion includes disc e e i s o align wi h use
p e e ences o ecommenda ions om ano he sys em, a-
o comple e songs wi hin a gi en s yle, o ocus on pa s
wi h speci ic pedagogical in e es :
Ck(g, , τ) = 0when he c i e ion is me
1o he wise
4.4 Song/Pa Cohe ency and Full Songs
Fo each gui a is gand challenge alue τ, i s a e com-
pu ed agains all e sions (Figu e 3). The model selec s
e sions ha o e he bes i s, minimizing Fg( , τ). Ad-
di ionally, we ensu e ha a gi en song and pa appea in
only one challenge alue o a oid edundancy.
Full songs (bo om lines on Table 2) can also be sug-
ges ed, in di e en e sions. They a e e alua ed based on
kαkβkγk,τ
challenge alue τ1 2 3 4
CCho ds 0.5 1 −0.5 0 1 3
PRhy hmic Pa e ns 0.5 1.5 −0.5 0 1 3
TTempo/Impac . 0.5 1.5 −0.5 0 1 3
GGlobal 0.8 0.3 −0.5 0.5 1 3
EExe cises Numbe 2 0.8 2 3 5 7
Table 3. Coe icien s o each c i e ion. The γ alues we e
se inc easing on he 4 challenges. Challenge 2 is mean o
be almos he es ima ed le el o he gui a is .
he di icul y o hei mos challenging pa s – occasion-
ally adjus ed o accoun o addi ional complexi y. Full
songs a e gi en a bonus in he anking p ocess, encou -
aging lea ne s o engage wi h comple e pieces.
4.5 Op imizing Weigh s
The coe icien s γk,τ we e chosen o a ge di e en chal-
lenges and o con ol he numbe o new exe cises in o-
duced (Table 3). The coe icien s αkand βkwe e e ined
du ing he e alua ion p ocess (see Sec ion 5.2). The small
amoun o da a a ailable, combined wi h he in en ion o
main ain a e y simple and in e p e able model, led us
o a o manual uning o e au oma ed op imiza ion ech-
niques, e en i i migh in oduce bias and o e i ing.
5. IMPLEMENTATION AND EVALUATION
5.1 Implemen a ion, Code and Da a A ailabili y
Comme cial applica ion. The Gui a Social Club applica-
ion, cu en ly unde de elopmen , p o ides gui a is s wi h
a comp ehensi e en i onmen cen e ed a ound gui a play-
ing, wi h mo e han 3200 ins uc ional ideos. Use s indi-
ca e hei p e e ed s yles, he songs hey al eady know,
and he songs hey wish o lea n. The model desc ibed
he e, wi h addi ional p op ie a y c i e ia, sugges song/pa
e sions, ca ego ized in o ou di e en challenge le els.
Gui a is s hen selec he piece hey wan o wo k on and
p og ess h ough a se ies o exe cises suppo ed by ideos
o inc easing di icul y. Videos may be sha ed ac oss mul-
iple pa s (e.g., eaching a speci ic cho d o cho d an-
si ion) o ailo ed o a pa icula pa (e.g., o a unique
hy hmic pa e n). The applica ion also acks which exe -
cises he use has comple ed, and his da a could be used
o enhance pe sonaliza ion.
Open-sou ce and open-da a componen s. De eloped in
Py hon, he code o he model desc ibed he e along wi h
i s e alua ion is eleased unde he LGPL 3+ license in a
gi eposi o y a ailable om algomus. /code. Al-
hough he comple e da ase is p op ie a y, we elease a
subse o his da a unde an open license, co e ing di -
icul y me ada a o 40 songs, 110 pa s, and 319 e -
sions, as well as 148 o acle e alua ions ela ed o hese
songs/pa s/ e sions (see below). This public da ase
is a ailable wi h he code as well as on he long- e m
eche che.da a.gou . a chi e a h ps://doi.o g/
10.57745/1KXHMJ .
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
300
G1 G2 G3 G4 G5 G6 G7 G8
0.3 0.3
0.2
0.2
0.5
0.5
0.5
0.6
0.6
0.7 0.2 0.5
0.3 0.2
0.2 0.2 0.3
0.2
0.2
Labels
- -3 -2 -1 0 1 2 3 +
a)
G1 G2 G3 G4 G5 G6 G7 G8
0.2
0.3 0.3
0.2
0.2
0.5
0.4
0.6
0.5
0.6
0.6 0.2 0.5
0.3 0.2
0.3
0.2
0.2
Labels
- -3 -2 -1 0 1 2 3 +
b)
G1 G2 G3 G4 G5 G6 G7 G8
0.2 0.2 0.2
0.2
0.2
0.3
0.2
0.2
0.5 0.3 0.7
0.2
0.3
0.8
0.5
0.4
0.3
0.3
0.5
0.2
0.3 0.2
0.2 0.2
Labels
- -3 -2 -1 0 1 2 3 +
c)
G1 G2 G3 G4 G5 G6 G7 G8
0.2
0.5
0.2
1.0
0.7
0.2
0.6
1.0
0.8 0.8
1.0
0.3
0.2
0.2
Labels
- -3 -2 -1 0 1 2 3 +
d)
G1 G2 G3 G4 G5 G6 G7 G8
0.2
0.2
0.2
0.2 0.3
0.6 0.5
0.6
0.4
0.6
0.8
0.3
0.5
0.2
0.2
0.3
0.3
0.4 0.2
Labels
- -3 -2 -1 0 1 2 3 +
G1 G2 G3 G4 G5 G6 G7 G8
0.3 0.3
0.2
0.2
0.5
0.5
0.5
0.6
0.6
0.7 0.2 0.5
0.3 0.2
0.2 0.2 0.3
0.2
0.2
Labels
- -3 -2 -1 0 1 2 3 +
Figu e 4. Dis ibu ion on de ia ions be ween he p edic ed challenge and he expe anno a ion on 77 ecommenda ions
om eigh es p o iles. (Le ). Full model. In a e age, 49% o he model sugges ions pe ec ly ma ch he challenge alue
om he expe (0), and 82% de ia e wi h a mos 1 challenge alue (+1 o −1). (Righ ). Abla ion models, igno ing a)
cho ds, b) PIPM and Rhy hmic Pa e ns, c) global di icul y, and d) numbe o exe cices.
5.2 E alua ion
We c ea ed eigh es p o iles G1−G8, ep esen ing di -
e en gui a is skill le els spanning om absolu e begin-
ne s (G1) o ad anced (G8). Fo each such gui a is g, he
model sugges s up o 12 song/pa e sions ha mini-
mize Fg( , τ)(3 pe challenge), wi h hus, o each sug-
ges ion , a sugges ed challenge φ( , g)∈[1,2,3,4]. The
gui a eache e alua es he ele ance o hese sugges ions,
p o iding o each sugges ion o a gui a is gan o acle
alue ψ( , g)∈[−∞,1,2,3,4,+∞]. The in ege s ep-
esen he expec ed challenge o he song/pa e sion, A
alue −∞ means he sugges ion is oo simple o gand
should ne e ha e been p oposed, while +∞indica es ha
i is a oo di icul . Du ing he op imiza ion o he coe -
icien s (see Sec ion 4.5), he expe gui a eache ac ually
labeled up o 720 sugges ions (including 77 e alua ions on
he inal sugges ions). This in o ma ion may be used o
u u e model op imiza ion and alida ion. Fo example,
on he gui a is p o ile G4, he ( ull) song 1. Knocking on
Hea en’s Doo (see Table 2) is anno a ed in he o acle as
ψ(1.c, G4) = 1,ψ(1.d, G4) = 2, and ψ(1.e, G4) = 3.
The Figu e 4 shows he dis ibu ion o ∆( ) = φ( )−
ψ( )on he 77 e alua ions o a ious models. Wi h he
coe icien s shown on Table 3, 82% o he sugges ions a e
conside ed pe inen by he eache , ha is wi h |∆| ≤ 1.
Abla ion models, in which ce ain c i e ia a e omi ed, s ill
achie e solid esul s, wi h a e age ele an p edic ion a es
anging om 60% o 100%. This highligh s bo h he o-
bus ness o he ull model and he signi icance o indi id-
ual c i e ia, pa icula ly he assessed o e all di icul y.
The obse ed e o s mos ly conce n songs ha a e ei-
he oo simple o oo complex, e en wi hin challenge
le els 1 and 4, espec i ely. Fu u e wo k will ocus on
s udying hese edge cases in g ea e de ail o imp o e he
model’s accu acy and ele ance.
6. DISCUSSION AND CONCLUSION
Tools suppo ing pedagogy and o e ing sugges ions
can help p e en lea ne demo i a ion du ing sel -s udy
phases [4]. They can also assis educa o s in ecommend-
ing sui able pieces wi h less p epa a o y wo k [12], p o-
mo ing independen and pe sonalized lea ning [15]. Sug-
ges ions should occasionally encou age lea ne s o s ep
ou side hei com o zone, as eache s o en ad ise.
Building on [13], we in oduced c i e ia o assess
hy hm gui a di icul y in Wes e n popula music and a
model o e alua e gui a is skill le els. To mi o eal lea n-
ing, we di ided songs in o pa s wi h mul iple e sions
linked o exe cises. This model is implemen ed in he Gui-
a Social Club applica ion, and we openly elease a pa
o his da ase . E alua ion agains expe o acle da a shows
ha 82% o he sugges ions a e ele an , alida ing his
ule-based sugges ion app oach as a i s p oo -o -concep .
The de e minis ic ule-based sys em and i s manual op-
imiza ion may ha e p oduce o e i ing, bu he ex emely
simplici y o he model (only 10 coe icien s αand β) e-
duced his isk. In any case, using an au oma ed sys em o
sugges new songs o lea n has limi a ions and is akin o
biases. De e mining which songs a lea ne knows p esen s
di icul ies – use s migh claim o know a song while lack-
ing he p o iciency assumed by he model. A gui a eache
could de ec such disc epancies.
In es iga ing how lea ne au onomy is balanced wi h
pedagogical guidance, and how pe sonaliza ion ex ends
beyond sel -decla ed mas e y, could enhance such me h-
ods. Fu u e wo k could explo e non-in usi e ways o e -
i y lea ne p o iciency, such as analyzing iming accu acy,
no e o cho d co ec ness a speci ic ins an s, o compa -
ing pe o mance pa e ns agains expe models. I should
include some andomisa ion o be e coun e -measu es o
limi biases [16] and be e e alua ion wi h ac ual use da a.
Finally, he amewo k p esen ed he e could also be ap-
plied o o he ins umen s o e en o ields beyond music
whe e skills and lea ning con en need o be ma ched.
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
301
7. ACKNOWLEDGMENTS
The au ho s hank he Algomus eam o hei aluable
eedback and he anonymous e iewe s, whose con ibu-
ions helped imp o e his pape . This p ojec is sup-
po ed by La Plaine Images, Hau s-de-F ance Inno a ion
Dé eloppemen , Bpi ance, he F ench Tech ini ia i e, and
he ANR TABASCO p ojec (ANR-22-CE38-0001).
8. ETHICS STATEMENT
While he eigh es p o iles we e c ea ed by he expe gui-
a eache wi h ce ain s uden s in mind, no pe sonal da a
was used in his s udy. The s udy does no in ol e any use
eedback o di ec use in ol emen . Fu u e esea ch could
include such aspec s, which would equi e p io e hical
conside a ion and app o al om he ele an e hics com-
mi ee.
9. REFERENCES
[1] R. C. Rod iguez and V. Ma one, “Gui a lea ning, ped-
agogy, and echnology: A his o ical ou line,” Social
Sciences and Educa ion Resea ch Re iew, ol. 8, no. 2,
pp. 9–27, 2021.
[2] A.-M. Bu ns, S. Bel, and C. T aube, “Lea ning o play
he gui a a he age o in e ac i e and collabo a i e
Web echnologies,” in Sound and Music Compu ing
Con e ence (SMC 2017), 2017.
[3] D. A senaul , “Gui a He o:" No like playing gui a a
all"?” Loading..., ol. 2, no. 2, 2008.
[4] M. Ma goudi, M. Oli ei a, and G. Waddell, “Game-
Based Lea ning o Musical Ins umen s: A Re iew
and Recommenda ions,” in Eu opean Con e ence on
Games Based Lea ning (ECGBL 2016), 2016.
[5] R. Rod iguez and V. Ma one, “Lea ning beyond he
game: A mul imodal analysis o Rocksmi h use s in-
e ac ions,” Ac a Ludologica, ol. 3, no. 2, pp. 36–53,
2020.
[6] J. Ma in-Gu ie ez, M. S. Del Rio Gue a, V. Lopez-
Chao, R. H. So o Gas elum, and J. F. Valenzuela Bo-
jó quez, “Augmen ed eali y o acili a e lea ning o
he acous ic gui a ,” Applied Sciences, ol. 10, no. 7,
p. 2425, 2020.
[7] M. Ba he , A. Anglade, G. Fazekas, S. Kolozali, and
R. Mac ae, “Music ecommenda ion o music lea n-
ing: Ho abs, a mul imedia gui a u o ,” in Wo kshop
on Music Recommenda ion and Disco e y (WOMRAD
2011), 2011, pp. 7–13.
[8] A. Xambó, J. Pauwels, G. Roma, M. Ba he , and
G. Fazekas, “Jam wi h Jamendo: Que ying a la ge mu-
sic collec ion by cho ds om a lea ne ’s pe spec i e,”
in Audio Mos ly on Sound in Imme sion and Emo ion
(AM 2018), 2018, pp. 1–7.
[9] S. A iga, M. Go o, and K. Ya ani, “S umme : An in-
e ac i e gui a cho d p ac ice sys em,” in IEEE In e -
na ional Con e ence on Mul imedia and Expo (ICME
2017), 2017, pp. 1057–1062.
[10] V. Sébas ien, H. Ralambond ainy, O. Sébas ien, and
N. Con uy , “Sco e Analyze : au oma ically de e min-
ing sco es di icul y le el o ins umen al e-Lea ning,”
in In e na ional Socie y o Music In o ma ion Re-
ie al Con e ence (ISMIR 2012), 2012, p. 571.
[11] P. Ramoneda, N. C. Tame , V. E emenko, X. Se a, and
M. Mi on, “Sco e di icul y analysis o piano pe o -
mance educa ion based on inge ing,” in IEEE In e -
na ional Con e ence on Acous ics, Speech, and Signal
P ocessing (ICASSP 2022), 2022.
[12] P. Ramoneda, J. J. Vale o-Mas, D. Jeong, and X. Se a,
“P edic ing pe o mance di icul y om piano shee
music images,” in In e na ional Socie y o Music In-
o ma ion Re ie al Con e ence (ISMIR 2023), Sep.
2023.
[13] M. A. V. Vásquez, M. Baelemans, J. D iedge ,
W. Zuidema, and J. A. Bu goyne, “Quan i ying he
ease o playing song cho ds on he gui a ,” in In e -
na ional Socie y o Music In o ma ion Re ie al Con-
e ence (ISMIR 2023), 2023.
[14] D. Régnie , N. Ma in, and L. Bigo, “Iden i ica ion o
hy hm gui a sec ions in symbolic abla u es,” in In e -
na ional Socie y o Music In o ma ion Re ie al Con-
e ence (ISMIR 2021), 2021.
[15] T. Vaizman and G. Ha paz, “Re uning music each-
ing: Online music u o ials p e e ences as p edic o s
o ama eu musicians music sel -e icacy in in o mal
music lea ning,” Resea ch S udies in Music Educa ion,
ol. 45, no. 2, pp. 397–414, 2023.
[16] A. Fe a o, X. Se a, and C. Baue , “B eak he loop:
Gende imbalance in music ecommende s,” in Con e -
ence on Human In o ma ion In e ac ion and Re ie al
(CHIIR 21), 2021, pp. 249–254.
P oceedings o he 26 h ISMIR Con e ence, Daejeon, Ko ea, Sep embe 21-25, 2025
302
