Voice Sculp o :
A ool o imp o ing public speaking abili ies
Analizado de Voz:
Una he amien a pa a ap ende a habla en público
T abajo de in de g ado del G ado en Ingenie ía In o má ica
Facul ad de In o má ica
Au o es
Guille mo O eje o Sánchez
Miguel Fe e as Chumillas
Di ec o es
Bo ja Mane o Iglesias
Jaime Sánchez He nandez
Cu so 2020–2021
Analizado de Voz
Una he amien a pa a ap ende a habla en público
Memo ia que se p esen a pa a el T abajo de Fin de G ado
Guille mo O eje o y Miguel Fe e as
Di igido po Bo ja Mane o Iglesias1y Jaime Sánchez He nandez2
1Depa amen o de Ingenie ía del So wa e e In eligencia A i icial
2Depa amen o de Sis emas In o má icos y Compu ación
Facul ad de In o má ica
Uni e sidad Complu ense de Mad id
Mad id, 2021
Abs ac
Speech is he main way we ha e o communica e o o he people wha we wan , eel o
hink. Howe e , and despi e he impo ance ha his has o socie y, he e is a la ge pa
o he popula ion wi hou enough p epa a ion o speak in public o wi h ea o doing
so.
The e a e di e en ways o imp o e a speech, be i he s uc u e, he message, o he
ocabula y used. This wo k ocuses on he sound quali ies o he oice, since hey can
be measu ed and analyzed by so wa e.
The ou come o his p ojec is a ool capable o ex ac ing measu able pa ame e s om
he oice - pi ch, in ensi y, speed and pauses - quan i ying and d awing ela ions be ween
hese me ics and p o iding eal ime da a.
To demons a e he applicabili y o ou so wa e we ha e ca ied ou wo expe imen s
based on he analysis o he speech o wo ypes o speake s, on one hand, p o essionals
used o public speaking and, on he o he , o a o y s uden s. In bo h cases, ou ool
has p o ed o be e ec i e in p ecisely ex ac ing oice p ope ies and a ia ions ha
de e mine he quali y o he speech.
Keywo ds
Audio, Real Time, Analysis, Voice, P aa , Pa selmou h
ii
Resumen
El discu so es el medio p incipal del que disponemos pa a comunica a o as pe sonas
lo que que emos, sen imos o pensamos. Sin emba go, y a pesa de su impo ancia en la
sociedad, hay una g an pa e de la población sin la p epa ación su icien e pa a habla en
público o con miedo a hace lo.
Exis en di e en es o mas de mejo a un discu so, ya sea la es uc u a, el mensaje o el
ocabula io u ilizado. Es e abajo se cen a en las cualidades sono as de la oz, ya que
se pueden medi y analiza a a és de so wa e.
El esul ado del p oyec o es una he amien a capaz de ex ae pa áme os medibles de
la oz - ono, in ensidad, elocidad y pausas - cuan i icando y azando elaciones en e
es as mé icas y p opo cionando los da os en iempo eal. Es os da os se p esen an en
una in e az de usua io que los isualiza en un o ma o legible, di igido a p o esionales
del discu so.
Pa a demos a la aplicabilidad de nues o so wa e, hemos ealizado dos expe imen os
basados en el análisis del discu so de dos ipos de o ado es: po un lado p o esionales
acos umb ados a habla en público y po o o es udian es de o a o ia. En ambos casos,
nues a he amien a se ha demos ado e icaz pa a ex ae de o ma p ecisa las p opiedades
de la oz y las a iaciones que de e minan la calidad del discu so.
Palab as cla e
Audio, Tiempo Real, Analisis, Voz, P aa , Pa selmou h
iii

Glossa y
ampli ude he maximum displacemen o dis ance mo ed by a poin on a ib a ing
body o wa e measu ed om i s equilib ium posi ion. I is equal o one-hal he
leng h o he ib a ion pa h. The ampli ude o sound is expe ienced as he loudness
o sound.
decibels o dB o sho is a uni o measu e he in ensi y o a sound o he powe le el
o an elec ical signal by compa ing i wi h a gi en le el on a loga i hmic scale.
o man s each o se e al p ominen bands o equency ha de e mine he phone ic
quali y o a owel.
equency is he a e pe second o a ib a ion cons i u ing a sound wa e, equency is
ep esen ed by cycles pe seconds and i ’s measu ed in He z (Hz).
in ensi y he a ia ion o he ene gy lux p oduced by he acous ic pe u ba ion.
pa alanguage is he non-lexical componen o communica ion by speech, o example
in ona ion, pi ch and speed o speaking, hesi a ion noises, ges u e, and acial ex-
p ession.
pa alinguis ic he s udy o he pa alanguage.
pi ch is he quali y o a sound go e ned by he equencies p oducing i ; he deg ee o
highness o lowness o a one ha can be pe cei ed by humans .
sound p essu e le el is he a io o he absolu e sound p essu e agains a e e ence
i
UCM
le el o sound in he ai .
speech a e is he speed a which a pe son speaks, i can be measu ed in wo ds pe
minu e o syllables pe second.
syllable is a uni o p onuncia ion ha ing one owel sound, wi h o wi hou su ounding
consonan s, o ming he whole o a pa o a wo d.
imb e is he pe cei ed quali y o a sound ha makes dis inc i e a pa icula oice o
ins umen .
one a musical o ocal sound wi h e e ence o i s pi ch, quali y, and s eng h.
ocal cue any meaning ul a ia ion in he sound o he oice du ing alk. These include:
ocal quali ie s ( a e, hy hm, du a ion, pi ch, one, a icula ion, loudness, pauses);
ocaliza ions; and ocal cha ac e ize s (laughing, c ying, yawning, coughing, and
so on).
wa wa e o m Audio ile o ma .
Sob e T
EF
L
O
NX
Te lon X(cc0 1.0(documen ación) MIT(código))es una plan illa de L
A
T
EX
c eada po Da id Pacios Izquie do con echa de Ene o de 2018. Con
a ibuciones de uso CC0.
Es a plan illa ue desa ollada pa a acili a la c eación de documen ación p o esional
pa a T abajos de Fin de G ado, T abajos de Fin de Más e o Doc o ados. La e sión
usada es la X
V:X O e lea V2 wi h XeLaTeX, ma gin 1in, bib
Con ac o
Au o : Da id Pacios Izquie o
Co eo: [email p o ec ed]
ASCII: [email p o ec ed]
Despacho 110 - Facul ad de In o má ica
i
Con en s
Página
1 In oduc ion 1
1.1 Mo i a ion ................................... 3
1.2 Goals ..................................... 3
1.3 Documen s uc u e ............................. 4
2 S a e o he a 5
2.1 So wa e o analyze speech .......................... 5
2.1.1 LikeSo App .............................. 5
2.1.2 O ai App ............................... 6
2.1.3 Voice Analys App .......................... 7
2.2 Audio and speech ............................... 8
2.2.1 In ensi y ................................ 9
2.2.2 Pi ch ................................. 10
2.2.3 Ha monics ............................... 11
2.2.4 Fo man s ............................... 12
2.2.5 Timb e ................................ 14
2.3 So wa e .................................... 14
2.3.1 Tools o analyze speech a ibu es ................. 14
2.4 Web F amewo ks ............................... 18
2.4.1 Flask ................................. 18
2.4.2 Dash .................................. 18
3 P ojec Implemen a ion 20
ii
UCM
1.1 Mo i a ion
Today, p ac ically all o us need o speak in public se e al imes h oughou ou li e. We
a e always alking and elling s o ies o people o di e en ages and all expe s ag ee ha
i we do no gi e a good speech, lis ene s will s op paying a en ion o wha we say o
hem.
In schools, public speaking is o en elega ed o he backg ound and w i ing ecei es
mo e a en ion, being much mo e s udied and p ac iced. Only in e y ew cases, schools
ac i ely each echniques ha imp o e ou abili ies o speak in public. This is he eason
why, in many cases, people a e a aid o gi e an o al p esen a ion, bu eel com o able
w i ing an essay.
Ou main mo i a ion o ca ying ou his p ojec has been o help people imp o e hei
speech and unde s and why hey usually ail o engage wi h hei audience. The objec i e
is o p opose an educa ional al e na i e o he lack o s udy on his o m o communi-
ca ion, while we ake he oppo uni y o deepen ou knowledge in he ields o public
speaking and audio in gene al.
1.2 Goals
The p ojec main goal is o de elop a ool capable o measu ing he quali ies o he oice
du ing a speech, p o iding eedback o he speake in eal- ime.
Sub-objec i es o he p ojec :
• S udy he quali ies o he oice, o include hose ha could be measu ed wi h
algo i hms and o be able o analyzing he oice main ea u es. (pi ch [30], in ensi y
[14], speech a e, pauses).
• Apply he ool o measu e changes in he oice o a g oup o s uden s a e an
o a o y cou se.
• Be able o de ec ypical e o s in a speake ’s pa alanguage.
• C ea e a epo wi h all he men ioned measu emen s.
3

UCM
1.3 Documen s uc u e
This documen is di ided in o 5 chap e s. The i s chap e is dedica ed o in oducing
he opic and p esen ing he mo i a ion and con ex o he p ojec .
In he second chap e , we explain he di e en audio ea u es we can ind in a speech and
we e iew some o he exis ing applica ions and ools o audio analysis.
The hi d chap e deals wi h he p ocess we wen h ough o c ea e he applica ion,
showing as well he ools we used.
In he ou h chap e , we de ail he expe imen a ion ca ied ou wi h he ool we de el-
oped, om he p epa a ion o he expe imen s o he esul s hey p oduced.
The i h chap e shows he conclusions we ha e eached a e he comple ion o he
p ojec and e e s o u u e wo k ha could be done in ou applica ion o imp o e i .
The six h and las chap e depic s ou indi idual con ibu ions o he p ojec .
Finally, he appendix co e s an ins alla ion me hod and a use guide o he applica-
ion.
4
Chap e 2
S a e o he a
He e will be discussed di e en so wa e and ools ha a e ele an o his p ojec .
2.1 So wa e o analyze speech
In his sec ion i will be alked abou simila so wa e ha ies o achie e one o mo e
o ou goals. Ano he poin ha will be discussed a e he simila i ies and di e ences
be ween wha has been achie ed and his so wa e.
2.1.1 LikeSo App
LikeSo is a mobile app designed o analyze a speech and gi e a eedback o imp o ing
i . I can analyze speeches up o 30 minu es long and he epo i gi es is ocused on
he ille wo ds and pace o speech (see Figu e 2.1). I also s o es all you epo s and
g ades, o show you p og ess. This app is only a ailable o iOS [22].
5
UCM
Figu e 2.1: Like So App
This app doesn’ ea u e a eal ime op ion and he ee op ion is e y limi ed. I ships a
ea u e o gi ing a sco e o a alk bu he documen a ion on why ha sco e is gi en is
no clea and don’ ake in conside a ion he numbe o pauses o du a ion o hem. Ou
app gi es mo e in o ma ion abou he speech bu doesn’ gi e a sco e o he use .
2.1.2 O ai App
O ai is a mobile applica ion simila o he p e ious one, bu wi h added ea u es. I gi es
he use eedback abou he ille wo ds an pace, bu in addi ion akes in o accoun he
in ensi y o he oice, he conciseness o he speech which is calcula ed in pa om he
numbe o wo ds ha a e epea ed. On op o ha allows o eco d a ideo o analyze
you acial exp ession while speaking [29], some hing ou o scope o ou p ojec .
6
UCM
Figu e 2.2: O ai App
This app is mo e comple e in e ms o he di e en me ics i gi es. Bu as well as he
o he s lacks he suppo o a eal ime analysis o he oice. The epo i gi es (see
Figu e 2.2) is e y simila o wha can be achie e wi h ou app bu mo e s ylized and
wi h be e explana ions and wha is good o bad and why. Addi ionally i comes wi h
a ille wo d ecognize .
2.1.3 Voice Analys App
This applica ion is he only one in his sec ion ha can gi e a eedback in eal ime [41].
I shows he pi ch le el and olume o he use while speaking, and some me ics ela ed
o hem, as he maximum, minimum, a e age and ange (see Figu e 2.3).
7
UCM
Figu e 2.3: Cap ion
This app comes also wi h a comme cial license. The app only measu es pi ch and in-
ensi y. Acco ding o hei web page, i ’s mo e ocused on medical pu poses a he han
o a o y. These app is simila o wha we achie ed bu wi h less me ics o gi e.
2.2 Audio and speech
In his sec ion we will in oduce di e en audio ea u es om speech ha will be use ul up
ahead. The ones desc ibed in he ideo om Emma Rode o [37] a e he mos impo an
o speech. This ea u es also ha e in common ha can be measu ed wi h compu e
so wa e and a en’ subjec i e.
8

UCM
2.2.1 In ensi y
When we e e o in ensi y, we e e a he measu emen s o changes in sound p essu e
le els measu ed in decibels (dB SPL) [30,33] as his is he way P aa measu es his
ea u e.
A de ini ion gi en by SCENIHR is [38]:
One pa ame e o he acous ic (sound) wa e which is gene ally used o assess
sound exposu e o humans is he sound p essu e le el exp essed in 𝜇𝑃𝑎 o
𝑃𝑎. Human ea audible sound p essu e le els ange om 20 𝜇𝑃𝑎 (hea ing
h eshold) ill 20 𝑃𝑎 (pain h eshold), esul ing in he scale 1:10,000,000.
Since using such a la ge scale is no p ac ical, a loga i hmic scale in deci-
bels (dB) was in oduced which is also in ag eemen wi h physiological and
psychological hea ing sensa ions.
As we can see in Figu e 2.4, he sound in ensi y ha P aa shows is gi en in dB (shown
below in g een), his le el o dB co ela es wi h he ampli ude [5] o he sound wa e
(shown abo e wi h bo h channels o he wa e o m), he b oade he signal he highe he
in ensi y cap u ed is.
Figu e 2.4: In ensi y ex ac ed om P aa
9
UCM
2.2.2 Pi ch
Pi ch is one o he main p ope ies o sound ha can be pe cei ed. I is he sound quali y
mos ela ed o equency which i ’s measu ed in He z (Hz). 1 He z equals 1 cycle pe
second. On a e age people can hea equencies be ween 20 and 20.000 Hz [42]. Humans,
on a e age ha e a Fundamen al equency (𝑓0)(also called one) a ound 100 o 120 Hz
o men and a ound 200 o 220 Hz o women [12].
Pi ch as de ined in B i anica [6]:
“Pi ch, in speech, he ela i e highness o lowness o a one as pe cei ed by
he ea , which depends on he numbe o ib a ions pe second p oduced by
he ocal co ds.”
Pi ch and equency a e bo h ela ed o each o he bu no in a linea way as we can see
in Figu e 2.5 ( equency plo ed in loga i hmic scale) equencies be ween 20 and 1000
Hz inc eases as e han o alues g ea e han 1000 Hz. Pi ch is he ea u e ha makes
humans pe cei e sound as ‘high’ o ‘low’ [50].
Figu e 2.5: Pi ch s F equency [42]
The meaning o ‘high’ and ‘low’ is ela i e o he imes pe second he wa e oscilla es
(see Figu e 2.6). The mo e slow i ib a es, he ‘lowe ’ i will sound.
10
UCM
Figu e 2.6: Pi ch compa ison
F om now on, we will use he e ms pi ch o one in e changeably, as meaning o say
undamen al equency (𝑓0). Fo wha ega ds o his p ojec gi en ha he me ic ha
mos in e es us is he ange o undamen al equencies he speake uses in a speech and
no he absolu e alue i has.
2.2.3 Ha monics
Ha monics a e a se ies o equencies de i ed om he undamen al equency. An ha -
monic is a posi i e in ege mul iple o 𝑓0. This undamen al equency is he loudes one
and also he one wi h he lowes equency [48]
11
UCM
Figu e 2.7: Ha monics (Daniel Bowling, 2010)
In bo h o he examples om he Figu e 2.7 we see a sound spec um and below ha he
ha monics ha o m ha sound. Ma ked wi h a black a ow can be seen he undamen al
equency ha s a s a 150Hz (in case o Figu e A), nex ha monics by de ini ion a e
a mul iple o he undamen al, so nex ha monics would con inue 300Hz, 450Hz, 600Hz
and so on. In addi ion i can be seen he i s wo ha monic peaks (F1 and F2).
2.2.4 Fo man s
A de ini ion o o man [30] would be he one in he P aa guide [51]:
A o man is a concen a ion o acous ic ene gy a ound a pa icula equency
in he speech wa e. The e a e se e al o man s, each a a di e en equency,
oughly one in each 1000Hz band. O , o pu i di e en ly, o man s occu
a oughly 1000Hz in e als. Each o man co esponds o a esonance in he
ocal ac .
Fo man s come om he ocal ac and depending on whe e and how he ocal ac
esona es ( ongue, la ynx) he ib a ion p oduced by hese will be di e en . In esul o
ha i will make a di e en equency o he o man s [48]. Dis inguishable o man s
usually ange om 𝐹1 o 𝐹4, bu some owels also expand up o 𝐹6, hese o man s, as
he quo e abo e poin s ou , goes in oughly 1000Hz bands.
As we can see in Figu e 2.8, he e a e six dis inguishable o man s bands ac oss he audio,
12
UCM
comes wi h some limi a ions due o he lack o good documen a ion he e a e some edge
cases ha a e difficul o ob ain. Dash abs ac mos o he code o building a web
app and c ea ing an in e ac i e igu e, so i s easy and quickly o i e a e and y di e en
op ions. I also ea u es a ho eload ea u e ha eloads he app wi hou he need o
s op i . Dash was used in he de elop o bo h epo s, in he li e epo used no only as a
isualiza ion lib a y, bu also as a se e o upda e he da a o hose isualiza ions.
19

Chap e 3
P ojec Implemen a ion
In his chap e will be co e ed he p ojec de elopmen and implemen a ion o he p o-
g am.
3.1 P og amming language
When selec ing a p og amming language o de elop his applica ion we picked up Py hon.
The main eason o pick Py hon o e any hing else (R, Julia, C#) was ha e en he e a e
o he p og amming languages used o scien i ic pu poses and wi h lib a ies o manipula e
audio, Py hon is in ac he one wi h he la ge communi y and wi h mo e documen a ion
and lib a ies a ailable o ou use case. We also we e amilia wi h Py hon so i was also
ano he key poin in a o o i . The las ac o was ha Py hon allows us o y di e en
hings quickly and o change and adap o new unc ionali ies ha we wan ed o add o
he applica ion.
3.2 P og am s uc u e
The s uc u e o he code was di ided in o co e unc ionali ies:
• Li e Repo : applica ion ha se es a web se e wi h Dash o eal ime analysis.
• De e ed Repo : applica ion ha uns a Flask se e o se e he webpage.
20
UCM
• U ils: ha ing code in common wi h some o he unc ionali ies.
• Analysis: Jupy e No ebooks iles used o y di e en hings in a isual way.
In Figu e 3.1 we can see 3 main componen s o he de e ed epo . The py hon com-
ponen which p ocess he audio and e u ns he in o ma ion o he Flask se e so i can
ende he HTML and se e i o he clien .
Figu e 3.1: De e ed Repo A chi ec u e
The a chi ec u e o he li e epo (see in Figu e 3.2) is a bi di e en om he o he
one. Fi s , he communica ion doesn’ happen be ween he Flask se e and he py hon
p ocess ha eco ds and p ocess he audio. Bu a he independen ly, Py hon sa es he
in o ma ion o s o age and hen he Flask se e ob ains he las in o ma ion ha was
sa ed so i could upda e he iew o he clien .
Figu e 3.2: Li e Repo A chi ec u e
21
UCM
3.3 P og am de elopmen
In his sec ion will be discussed he p og am de elopmen and implemen a ion o i .
Fo a i s d a o he p og am we conside ed a good s a ing poin he me ics ha
Emma Rode o highligh s in he TED alk, “Pe suade con u oz. Es a egias pa a sona
c eíble” [37].
The ou me ics Emma speaks abou in he alk a e he ollowing:
a). In ensi y
b). Pi ch
c). Speed Ra e
d). Timb e
As imb e was a difficul hing o measu e and cap u e we subs i u ed his me ic wi h he
numbe o pauses, ano he impo an me ic in o a o y [19]. This pauses we e measu ed
using a P aa sc ip which will be discussed la e .
The isualiza ions in Figu e 3.3 we e made in an in e ac i e Jupy e No ebook om a
eco d o audio o 8 seconds o al. The da a on hese 3 g aphs, was ob ained wi h Pa sel-
mou h unc ions o ob ain his pa icula da a (ampli ude, in ensi y, equency).
22
UCM
Figu e 3.3: Wa e ampli ude, In ensi y and pi ch om a 8 seconds eco d sample
To see how much efficien Pa selmou h calls a e we also measu ed he calls o unc ions
om Pa selmou h in Figu es 3.4,3.5
23
UCM
Figu e 3.4: Pe o mance o calls ‘ o_in ensi y‘ and ‘ o_pi ch‘ om an audio wi h a
du a ion o 1.63s
Fo di e en audios o di e en leng h we can see ha he calls o he lib a y o he unc-
ion ‘ o_pi ch’ goes up in ime linea ly (see Figu e 3.5) wi h an algo i hmic complexi y
o 𝑂(𝑛) gi en ha 𝑛is he numbe o by es in an audio ile.
Figu e 3.5: Pe o mance o calls ‘ o_pi ch‘ wi h audios o di e en du a ion
24

UCM
3.4 De e ed epo
Al hough he main goal was o ge a eal ime ool ha could gi e eedback o he
use , a De e ed epo o he speech was also de eloped, his epo could ill ano he
pu pose o he use o he app which could gi e some eedback once he speech is done
o ge a e ospec i e iew o i , as you can hea again some po ions and ge a deepe
unde s anding o wha wen w ong. This epo gi en he audio ile i will p in he
s a is ics and g aphs o he audio he use passed as pa ame e s o he applica ion.
This epo is c ea ed in a web se e ha e u ns a HTML page ha could be highly
cus omized and se ed ia he in e ne , i was done his way because his could be
consumed as a ee So wa e as a Se ice (SaaS) on he in e ne so he use doesn’ e en
ha e o ins all any hing on hei compu e and jus uploading an audio ile o ge he
epo o he speech in a ew seconds.
This epo was buil on op o he Flask web se e amewo k wi h a empla e in HTML
ha would be ende ed using he empla e engine ha comes wi h Flask, Jinja2. This
ende ed HTML p in s some ables om a JSON e u ned om he ‘Syllable Nuclei’
sc ip s (see Figu e 3.6) and also plo wo g aphs om he da a con ained in NumPy
a ays. These plo s a e in e ac i e in a way ha pi ch and in ensi y could be zoomed in
and ou and also panned h ough he whole leng h o he audio.
25
UCM
Figu e 3.6: Table wi h audio in o ma ion
As a s a ic g aph wasn’ easy o ge insigh s om i , a mo e app op ia e app oach was o
plo his g aphs in an in e ac i e way, o his we make use o Dash g aphing capabili ies
o plo on he web he in ensi y and pi ch. As Dash is buil on op o Flask i ha es an
easy in eg a ion wi h he Flask web se e .
Figu es 3.7 and 3.8 a e made in e ac i e wi h Dash, as i has Ja asc ip lib a ies o
dynamic plo ing, so in he i s igu e (3.7) he whole audio in ensi y could be seen, bu
o see mo e clea some po ions o audio i could be selec ed and zoomed in like in Figu e
3.8
26
UCM
Figu e 3.7: In ensi y plo (Dash)
Figu e 3.8: In ensi y plo , selec ing a speci ic ime window o isualize i clea ly
Ano he impo an ea u e was o be able o playback he speech in he same epo . This
was made in he empla e including an HTML ag o audio ha is capable o playing
audio in he web b owse .
Figu e 3.9: Audio Con olle
27
UCM
3.4.1 T ansc ip ion
Exis s se e al me hods o audio ansc ip ion (also called Speech o Tex ). He e we will
lis some o hem and discuss he use o one o e he o he s.
As di e en me hods exis we will ocus on he ones ha uses machine lea ning models
o p edic he ex , as nowadays hese me hods gi e he mos accu a e esul s. We
dis inguished hem be ween wo di e en ypes. Cloud based o in-house.
An exis ing py hon lib a y ha ha e Speech o Tex models o using a cloud se ice om
AWS, Azu e o IBM. Since we wan ed o ge esul s ha we e close o eali y bu also
as and wi h pa ame e s ha could be con igu ed we op ed o use he cloud companies
since hese usually ha e be e ained models o Speech Recogni ion o a a ie y o
di e en languages, lib a ies wi h p eloaded models usually come wi h a single model
ha has been ained wi h mainly English da a, bu since we we e analyzing Spanish
speake s ha won’ be as p ecise as a Spanish ained model. This cloud se ice om
‘IBM Wa son Speech o Tex ’ also o e s a ee ie o y wi h 500 minu es pe mon h
o Speech o Tex compu a ions. The IBM Speech o Tex also comes wi h di e en
pa ame e s o une he se ice. One o his pa ame e s calcula es a imes amp o each
wo d o be e audio labeling.
In Figu e 3.10 we can see a ull example o wha IBM o e s in his se ice. I can de ec
mul iple speake s in an audio and add wo d imings.
28
UCM
Expe imen al design
Fo his i s expe imen we wan ed o y he accu acy o he eal ime analysis o ou
applica ion. To achie e i , we used a compu e wi h Linux Min 20.2 Cinnamon as he
ope a ing sys em ins alled. We downloaded a ool called Pa ucon ol o change he inpu
ha ou compu e ecei ed, and wi h he audio ou pu as he new inpu de ice, we an
ou applica ion and s a ed isualizing he ideos one by one.
Fo he analysis, we chose wo andom minu es om each speech. In each one, he i s
i een seconds we e dedica ed o illing he bu e o ou applica ion and he nex 45 o
eco d he a ia ion o he da a in eal ime.
Once he esul s we e eco ded, we compa ed hem wi h he s anda d o wha would be a
good speech. To ob ain his s anda d, we elied on he esea ch o Emma Rode o, whe e
she de ines he ideal pa ame e s as ollows [37]:
• In ensi y: High, wi hou shou ing, a ound 40 o 50 dB
• Pi ch: A low pi ch, his is ela i e because by na u e humans ha e di e en pi ches
(women no mally has highe pi ched oice han men)
• Speed: A ela i e good speed would be a ound 160 o 180 wpm. Which ansla es
o a ound 4 o 4.5 syllables pe second (since in Spanish a wo d has a mean o
a ound 1.7 syllables [15]).
Ano he impo an ac o o he e ec i eness o he speech is he numbe and du a ion o
he pauses, since pauses gi e ime o he audience o assimila e wha has jus been said.
Howe e , i he pauses consis on a ille wo d o hey a e oo long, hey can dis ac he
audience om he speech de elopmen . In a s udy om E. Rode o [36], a good numbe
o pauses would be a ound 10 pe minu e, ye his s udy was conduc ed wi h English
speake s du ing a adio bulle in. We would usually see mo e pauses pe minu e in a
common speech and we should also ake in o accoun he di e ences in languages. Thus,
Spanish speake s end o speak as e and o pause mo e du ing speech.
35

UCM
Figu e 4.1: Expe imen Design o ‘Anylyzing g ea speake s’
4.1.2 Resul s
Emma Rode o
Speech Video: Pe suade con u oz. Es a egias pa a sona c eíble
The ollowing minu es om he ideo we e selec ed o be analyzed: 2:00 - 3:00 and
6:00-7:00
Du ing he wo minu es eco ded, Emma Rode o main ains a cons an speed, which
d ops only when she elies on he isual suppo she has. Thus, he speech a e emains
cons an ly be ween ou and ou and a hal poin s, only dec easing o h ee and a hal
when he e is a longe pause. Besides, h oughou he alk, he speake elies e y li le
on pauses, keeping hem be ween one and h ee e e y 15 seconds.
Figu e 4.2: Emma Rode o speed sample
36
UCM
When we look a he pi ch we can see a pe ec use o pauses o inish he sen ences, as
shown in he igu e below, in addi ion o a e y wide ange o alues.
Figu e 4.3: Emma Rode o pi ch sample om he i s minu e
In he second minu e, Rode o speaks abou he pi ch. The e is a e y cha ac e is ic pa
in which, o ep esen how di e en pi chs a ec he speech, she changes he own while
alking, i s oo high, and hen oo low. The ollowing igu e depic s he pi ch shown in
he applica ion du ing his pa .
In he image you can clea ly see h ee s ages, he one on he le co esponding o when
he pi ch begins o inc ease, he one in he middle ep esen ing he pa wi h he lowes
pi ch, and he one on he igh he speake ’s no mal oice.
Figu e 4.4: Emma Rode o pi ch sample om he second minu e
As he da a shows, and as expec ed, his speech i s pe ec ly in o he s anda ds, main-
aining a cons an and adequa e speed, wi h no oo many pauses and a pe ec use o
he pi ch.
Rega ding he in ensi y, we can see ha he esul s shown by ou applica ion a e no
well calib a ed, since acco ding o he da a ob ained he in ensi y exceeds 200db a some
poin , some hing ha clea ly does no happen. On he o he hand, and despi e he ac
ha he absolu e alue is no co ec , he a ia ion can co espond o eali y.
37
UCM
Figu e 4.5: Emma Rode o in ensi y sample
Blanca Po illo
Speech Video: El ea o nos enseña que equi oca se no es un d ama
The ollowing minu es om he ideo we e selec ed o be analyzed: 1:00 - 2:00 and
3:00-4:00
Figu e 4.6: Blanca Po illo speech a e sample
I we lis en o his speech we can see ha he speake alks calmly and is no a aid o
pause. This is e lec ed in ou applica ion, which du ing he wo minu es analyzed shows
a numbe o pauses ha a ies be ween h ee and six, and a speech a e be ween h ee
and ou . I we compa e he speed wi h he canon, i is sligh ly slowe han i should be.
Howe e , when alking abou anecdo es and pe sonal expe iences, he speake p io i izes
a calm and slow one ha makes he message clea , compa ed o a as e one ha sounds
mo e con iden .
In he ollowing igu es we can see samples o he pi ch ex ac ed om he wo minu es
analyzed.
38
UCM
Figu e 4.7: Blanca Po illo pi ch sample om he i s minu e
Figu e 4.8: Blanca Po illo pi ch sample om he second minu e
As he da a shows us, he i s minu e mee s he s anda ds when i comes o dealing wi h
pauses, ending each sen ence in a low pi ch. Howe e , in he second minu e, we can see
how he speech begins o de ia e om he s anda d and he sen ences s a o end in he
same pi ch in which hey begin o e en in a highe one.
A he beginning o he second minu e, when Po illo is ecalling an anecdo e, one o he
sen ences ends on a much highe pi ch han he one she s a ed wi h, making he lis ene
eel ha he sen ence, ollowed by an uncom o able pause, is no o e . When we look a
ou applica ion, his is pe ec ly e lec ed as shown in he ollowing igu e.
Figu e 4.9: Blanca Po illo high pi ch a he end o a sen ence
Ma io Alonso Puig
Speech Video: En odo se humano hay g andeza
The ollowing minu es om he ideo we e selec ed o be analyzed: 19:00 - 20:00 and
39
UCM
38:00-39:00
In gene al, du ing he speech, Alonso Puig main ains an adequa e speed al hough his
a ies g ea ly depending on wha he is elling, he also in oduces a la ge numbe o pauses
ha inc ease he a ia ion. He has an a e age speech a e o a ound h ee and ou and
a hal , and his pauses ha e alues be ween wo and six al hough hese a y g ea ly.
Figu e 4.10: Ma io Alonso Speed sample
I we look a he pi ch, in he i s igu e we can see a cu ious echnique ha is qui e a
om he canon. I occu s on se e al occasions du ing he speech, and consis s o using
a la ge numbe o sho -medium pauses p eceded by sen ences ha end in a high pi ch.
On he con a y, in he second igu e we can see ha he speake also adap s his speech
o he s anda ds, using less pauses and inishing he sen ences in a low pi ch.
Figu e 4.11: Ma io Alonso pi ch sample om he i s minu e
40

UCM
Figu e 4.12: Ma io Alonso pi ch sample om he second minu e
Edga Cabanas
Speech Video: Las cla es pa a ende la elicidad
The ollowing minu es om he ideo we e selec ed o be analyzed: 4:00 - 5:00 and
15:00-16:00
In i s o m, his is a e y simila speech o he one o Ma io Alonso Puig. Du ing he
minu es analyzed, he a e age speech a e is highly a iable and i luc ua es be ween
h ee and a hal and ou and a hal poin s due o he la ge use o pauses. The eco ded
pauses also a y be ween wo and se en.
Figu e 4.13: Edga Cabanas speed sample
When we look a he pi ch, he simila i ies wi h Puig’s speech a e clea . Cabanas also
41
UCM
uses a la ge numbe o pauses – sho e , hough - inishing he p e ious sen ence in a
high pi ch. And, as Puig, he does no use his echnique du ing he whole speech bu ,
as we can see in bo h he i s and second igu es, mos o he ime he adap s o he
s anda ds.
Figu e 4.14: Edga Cabanas pi ch sample om he i s minu e
Figu e 4.15: Edga Cabanas pi ch sample om he second minu e
In he igu e below we a e able o see he eal ime in ensi y ou applica ion show us.
Du ing he wo minu es analyzed he e we e no mayo changes in he speake in ensi y,
who main ained a egula oice le el.
Figu e 4.16: Edga Cabanas in ensi y sample
He nan Cascia i
Speech Video: Mi hija quie e en ende el sis ema inancie o
The ollowing minu es om he ideo we e selec ed o be analyzed: 4:00 - 5:00 and
13:00-14:00
42
UCM
In his speech Casa i uses a slow hy hm, wi h a speech a e ha a ies be ween h ee
and ou poin i e due o he use o pauses. The la e a e equen ly used o emphasize
he sen ences.
Figu e 4.17: He nan Cascia i speed sample
In he analysis we can see how he speech does no ully adap o he s anda ds. The
speake o en lea es a high pi ch be o e a sho -medium pause o ge he lis ene ’s a en-
ion and hen he inishes he concep wi h a sen ence ended in a low pi ch. This can be
clea ly hea d in he ideo and is ep esen ed in he igu es below.
Figu e 4.18: He nan Cascia i pi ch sample om he i s minu e
43
UCM
Figu e 4.19: He nan Cascia i pi ch sample om he second minu e
4.2 Analyzing s uden s om an o a o y cou se
4.2.1 Me hodology
Pa icipan s
In his expe imen we had he pa icipa ion o ele en s uden s om di e en deg ees
wi hin he Complu ense Uni e si y o Mad id, nine women and wo men. All o hem
we e young people unde he age o hi y.
Expe imen al design
Fo his expe imen we had access o a h ee-day o a o y cou se a he Complu ense
Uni e si y o Mad id, whe e we eco ded wo speeches om each pa icipan .
The i s eco dings co espond o he i s speech gi en by he s uden s a he beginning
o he cou se on a gi en opic. The opic was abou he eason why hey conside ed
hemsel es good candida es o a ce ain posi ion a he uni e si y adminis a ion. To
p epa e his opic, he s uden s had en minu es.
In o de no o in luence he speech o make he pa icipan s ne ous we we e no allowed
o a end he e en , so he eco dings we e gi en o us by he cou se p o esso . The
eco dings we e made o bo h ideo and audio, being he audio he one collec ed wi h
he came a’s mic ophone.
The second eco ding session co esponded o he inal speeches o he cou se’s s uden s.
Fo hese speeches he s uden s chose hei opics and whe he o no hey wan ed a
44
UCM
main ained he same in ensi y in bo h eco dings. The e is also an o e all inc ease in
s anda d de ia ion, indica ing a wide use o olume le els by s uden s.
Looking a he main pi ch we can see a big di e ence be ween men and women. Men s a
om a pi ch be ween wo and h ee imes lowe han women and bo h sligh ly inc ease
i om he i s o he second alk. On he con a y, women dec ease i be ween wen y
and o y he z app oxima ely.
This di e ence is no ound when we look a he s anda d de ia ion. In gene al, in his
pa ame e we can obse e an inc ease om he i s speech o he second. This inc ease
has highly a ied alues, going om wo and a hal o hi y- h ee he z. The only
excep ion is s uden numbe h ee. Howe e , i we look a he o al alue and no a he
compa a i e one, he alue he achie es is e y simila o he one o o he s uden s in he
second speech.
This gene al inc ease implies ha in he h ee days o he cou se he s uden s lea ned o
widen he pi ch ange hey used, which allowed hei speeches o sound less mono onous.
In he ollowing pa ame e s we can see he du a ion - which we only use o calcula e he
a e ages o o he pa ame e s -, he ASD, ha ba ely changes, and he speed, ep esen ed
by he a icula ion and speech a e. Due o he low p ecision o he pauses, his las
ac o lea es us wi h unclea da a.
51

Chap e 5
Conclusions and u u e wo k
5.1 Conclusions
In his p ojec we s udied how pa alanguage in luences he a o public speaking o c ea e
an applica ion ha p o ides o a o s wi h eal ime and de e ed eedback on he quali y
o hei speech.
The expe imen s ca ied ou demons a e how ou applica ion is capable o showing
enough da a o unde s and he shape o a speech in eal ime.
In he i s expe imen , we obse ed how oice da a collec ed om ‘good’ speeches sha e
mul iple common ai s. Some o his ai s can also be linked o he pa alanguage
s anda ds p oposed by Emma Rode o, which de ine he cha ac e is ics o an e ec i e
message deli e y when speaking in public in Spanish language.
We could iden i y he ollowing ai s be ween p o essional speake s:
• They all use a as e speed when ying o ansmi a emo ion, and a slowe wi h
longe pauses when hey wan he audience o emembe a speci ic message.
• E en hough hey use di e en s a egies o he use o he pi ch o con on pauses,
hey all eso o he s anda ds many imes du ing he speech.
The second expe imen demons a es how ou applica ion can p o ide aluable da a o
52
UCM
ack p og ess in public speaking. We compa ed oice eco dings om s uden s in an
o a o y cou se, bo h be o e s a ing and a e concluding he cou se. We could obse e he
imp o emen s s uden s made jus by looking a he da a. This imp o emen s co espond
mainly o a wide and a mo e app op ia ed use o hei oice in ensi y and pi ch.
The expe imen s we e ca ied ou wi h ce ain limi a ions. In he i s , due o he ma e ial
used, we did no ge a good inpu o ce ain pa ame e s, wha ga e us un eal alues in
in ensi y making i no use ul o he analyses. In he second expe imen he backg ound
noise o he i s ba ch o eco dings al e ed he da a ela i e o speeds and pauses gi ing
us no alid da a o compa e.
5.2 Fu u e Wo k
Ou applica ion has been able o ex ac om a good inpu sou ce accu a e me ics abou
he in ensi y, pi ch, speed and pauses in a speech. The u u e wo k will consis on he
ollowing.
• Adding he sc ip we did o analyze he epo s in he expe imen , o immedia ely
ge me ics like mean and s anda d de ia ion o he pi ch and he in ensi y.
• In eg a ing in he eal ime epo he ansc ip ion, ocusing on he ille wo ds
and he numbe o hem. The ille wo ds a e a good indica ion o whe he a speech
is good o no , so adding hem will imp o e ou applica ion.
• Linking he ansc ip ion wi h he me ics g aphs in o de o be e unde s and
how ce ain alues a e ela ed o ce ain sen ences along he speech.
• Di iding pauses in di e en ypes, sho , medium o long, depending on he du-
a ion, because each one should be con on ed in a di e en way. Adding hese
ca ego ies and making ou ool able o show how a pe son uses he di e en pauses
can gi e e y aluable da a o he analysis.
• C ea e as e and smoo h upda es. In dash also exis clien side callbacks which
can be execu ed in in e als o less han 100 milliseconds bu needed he da a o
be s o ed in he clien . Implemen ing his will imp o e he eal ime analysis by
53
UCM
showing in o ma ion wi hou any delay.
• Modi y he eal ime applica ion in e ace o allow o see all he me ics a he same
ime, o no ha ing o epea he analysis o ocus on di e en me ics.
• Repea he expe imen s wi h a g ea e and mo e a ied numbe o pa icipan s,
and a ma e ial ha allows ou applica ion o ob ain all he pa ame e s in a alid
way.
We belie e he ool we ha e designed and he da a i shows can be e y use ul in di e en
ields o esea ch, applicable o mul iple indus ies. In psychology, o ins ance, his
applica ion could help o unde s and how ce ain s imuli a ec s a pe son’s speech. In
o a o y, he ou pu da a p o ided by he app could ack a pe son’s p og ess in lea ning
public speaking skills o i could be used o ge in o ma ion abou he quali y o a speech
in compa ison o a ce ain s anda d. These a e jus some o he possibili ies we ha e
hough abou bu , in he u u e, his applica ion could be used o de elop o he p ojec s
in nume ous ields.
54
Chap e 6
Indi idual wo k
6.1 Guille mo O eje o
I s a ed esea ching o ools ha could gi e us in o ma ion abou audio and mo e
speci ically he oice, he e a e a lo o comme cial ools ha make his job possible,
some a e open sou ce ones, and o he a e paid apps. We immedia ely disca ded he paid
apps and s ick wi h he ones ha we e ee o use. Mos o his ools also ea u e a lo
o complex ans o ma ions ha can be applied o audio, his could be an in e es ing
ea u e o p ep ocess some o he audio iles o educe noises. Bu since i was going o
be wi h eal ime audio. I didn’ bo he o esea ch abou hese complex ans o ma ion
ha a e compu a ionally hea y and equi e a lo o ime o inish.
The wo main ools I in es iga ed we e:
• P aa
• SonicVisualize (w/ Aubio Plugin)
These ools we e used mainly o isualize da a, bu hey ha e a wide ange o possibili-
ies.
The nex hing I s a ed de eloping was a way o ob ain o man s om an audio wi h he
pu pose o analysing a ea u e o imb e and also as a i s app oach o measu ing speed,
since i owels could be coun ed, i will be easy o ge he numbe o syllables. I eco ded
55
UCM
mysel saying he owels and de elop a sc ip ha checked, based on in o ma ion abou
i s and second o man anges, wha owel was being p onounced. A he end we don’
end up using his o he inal p ojec , bu i was an in e es ing hing o de elop and
lea n abou .
In he oppo uni y we ha e o a end some o he Bo ja’s o a o y classes he e was a lo
o ask o do o eco d he s uden s as and wi hou in e up ing hem. As he e was
mul iple ask o do. I was in cha ge o assu ing he came a and he mic ophone we e
eco ding a all imes and he e was no noise in he mic ophone due o ic ion wi h he
clo hes, also because I ca ied my lap op o e he e I was also in cha ge o sa ing he iles
once he class was o e o la e p ocessing and emo al o he image, so I la e could
upload hem o ha e access o hem when needed. These eco dings we e use ul la e on
when he applica ion was inished and he expe imen a ion phase s a ed.
The nex hing was o build a simple web app wi h lask ha gi en an audio ile ende ed
a HTML wi h he me ics. The i s ini ial e sion was e y simple and ba ely ha e any
s yling wi h CSS. This e sion was i e a ed o e o add mo e complex ables, s yle and
he in e ac i e g aphs.
One o he ea u es ha was also included in he de e ed epo was a ansc ip ion o
he ex . Miguel esea ched he di e en op ion ha we e a ailable o make his ea u e.
He old me abou he IBM one and decided o s ick wi h i . Then I s a ed es ing wi h
he SDK IBM has a ailable and implemen ed he ansc ip ion in he epo .
Finally he de eloping o he li e epo was di ided in wo pa s, he communica ion
p ocess and he dashboa d. I s a ed ying wi h di e en ways o communica e wi h he
dashboa d. The i s a emp s using a message b oke ool we e e y complex o he
p ojec . They a e pe ec ools o messaging wi h high pe o mance, as in his con ex
we didn’ need ha pe o mance we ied simple ways o communica e wi h p ocess, he
message b oke ools I used we e Redis, a In-memo y da abase wi h capabili ies o be
used as a message b oke . And Rabbi MQ, a ully unc ional message b oke ha comes
wi h mo e con igu a ion and ad anced op ions o se he b oke s han Redis.
As we sea ched o o he ways o communica e be ween p ocesses I de elop a couple o
p oo o concep s o communica ing wo py hon p ocesses wi h socke s, one sending audio
da a and he o he ecei ing he da a om he socke po . This idea was main ained o
56

UCM
a bi because i wo ked o e e y OS and also was as enough. This solu ion was good bu
a bi complex o debugging. And gi en he dash limi a ion o upda es a a minimum o 1
ame pe second wasn’ he bes op ion ei he . In he side o he dashboa d de elopmen
I in oduced some small ea u es like he use o he speed gauge and a play/s op bu on
o make he in e ace mo e use iendly. Also ela ed wi h he applica ions I made
some documen a ion so ha o he de elope s could imp o e he applica ion and a use
manual, o use s ha would like o y hese app bu don’ necessa ily ha e a backg ound
in compu e science.
6.2 Miguel Fe e as
When he p ojec s a ed, my main ask was o do esea ch on he opic, lea ning abou
audio, ocal cues, pa alanguage, and he e ec s o all hese ac o s on speech. This
esea ch helped us be e unde s and bo h he cha ac e is ics o speech and he me ics
we wan ed o measu e wi h ou app.
Du ing his ime, I was also analyzing exis ing applica ions ha we e close o he idea
we wan ed o de elop, in o de o see wha hey did, how hey wo ked and wha ype o
e o s we should a oid.
Once we go a p elimina y design o ou app, in addi ion o u he esea ch I es ed di -
e en ools and p og amming languages o ob ain he da a we wan ed o he oice.
The ollowing image co esponds o a small p og am ha I c ea ed in Uni y, which showed
in eal ime he in ensi y o he sound cap u ed h ough a mic ophone.
Figu e 6.1: Measu e o sound in ensi y in Uni y
57
UCM
A e ini ial esea ch, and once we decided we would use Py hon as a p og amming
language, I s a ed wo king wi h Jupy e No ebook o c ea e sc ip s capable o ex ac ing
basic me ics om a eco ded audio. To his end, I sea ched he in e ne o audio samples
ha we could use o see he accu acy o ou analysis. Howe e , i was no easy o ind
sui able samples wi h only one oice and no backg ound sound. To sol e his p oblem,
I c ea ed a small piece o code, which allowed me, hanks o he ‘pyaudio’ and ‘wa e’
lib a ies, o eco d my own oice o as many seconds as I wan ed and sa e i in a ‘.wa ’
ile. This made he ollowing p ocess much easie , as I was able o c ea e my own samples,
allowing me o modula e my oice and ge he one I wan ed.
By joining hese sc ip s wi h he wo k done by Guille mo, we c ea ed he basic s uc u e
o ou applica ion. Then I s a ed looking o ways o ge he p og am wo k in eal ime
as well.
Hal way h ough he p ojec , and hanks o Bo ja Mane o, we had he oppo uni y o
a end wo sessions o a public speaking cou se o eco d he s uden s’ speeches. Fo
hese eco dings, I was in cha ge o e iewing he ma e ial and going o each session ea ly
o p epa e and es i , making su e he e we e no p oblems when eco ding.
A e he cou se ended, we con inued wi h he de elopmen o ou app, di iding he wo k
o igu ing ou how o in oduce he missing me ics. Du ing ha ime, I disco e ed
he IBM Wa son ool ha Guille mo used o implemen he ansc ip ion o an audio in
ou app while I was in cha ge o inding a way o measu e and analyze he pauses o a
speech.
Wi h he app co e inished, I ocused on he sea ch o ools ha would allow us o c ea e
an in e ace capable o inco po a ing da a in eal ime, and I disco e ed Plo ly and Dash,
which we ha e used o he inal in e ace.
Finally, once he applica ion was comple ed, I was in cha ge o collec he necessa y da a
o he wo planned expe imen s, aking he me ics om bo h he YouTube ideos and
he audios eco ded du ing he public speaking cou se we a ended.
58
Appendix
Use Manual
Ins alla ion
Windows
P e e ed me hod o ins all Py hon is ia Chocola ey, a Windows package manage . Bu
could also be done wi h Anaconda
Ins all Py hon 3.8
Ins all Py hon 3.8 wi h p e e ed me hod
Ins all Dependencies
py −m ensu epip −−upg ade
pip i n s a l l wheel
pip i n s a l l PyAudio −0.2.11−cp38−cp38−win_amd64
pip i n s a l l − equi emen s . x
Unix
sudo ap −ge i n s a l l py hon3 .8
pip i n s a l l PyAudio −0.2.11−cp38−cp38−win_amd64
pip i n s a l l − equi emen s . x
59
UCM
Use Guide
De e ed Repo
En e o he “de e ed_ epo ” olde and un he ollowing command:
py hon epo _de e ed . py
Then access o localhos in po 5000:
h p://127.0.0.1:5000/
Upload a ile in .wa o .mp3 o ma and submi i .
This will ake you o ano he ou e like his:
h p://127.0.0.1:5000/ epo ? ile=my_uploaded_speech.wa
S eam Analyze
En e o he olde “s eam_analyze ” and un he ollowing command:
py hon main . py
Then access o localhos in po 5050:
h p://127.0.0.1:5050/
Click on he ‘play’ bu on o s a eco ding and s a analyzing you speech.
When done hi ‘s op’ and he eco d o ha session will be sa ed in he ‘ou pu ’ olde .
The app will pick you de aul mic ophone. I i doesn’ ecognize you mic ophone he e
is a sc ip in ‘u ils/ge _inpu _de ices.py’ ha would lis he di e en inpu de ices.
Con igu a ion
To change se ings o he s eam analyze app en e modi y he ‘.en ’ ile om he
‘s eam_analyze ’ olde and change he ollowing alues:
This a e he alues ha wo k ine by de aul .
60
“Always look on he b igh side o li e”
Mon y Py hon
Guille mo O eje o y Miguel Fe e as
2021
Las Upda e: Sep embe 21, 2021
L
A
TEX lic. LPPL & powe ed by T
EF
LO
NCC-ZERO
This wo k is licensed unde a C ea i e Commons “CC0 1.0
Uni e sal” license.