Figu e 1: A Li eLily session.
COMACROB - Ins an Syn hesis Th ough Li e
Coding and AI
Alexand os D ymoni is
Depa men o Design and Media
Cyp us Uni e si y o Technology
Limassol
[email p o ec ed]
Abs ac
COMACROB is he ac onym o he pos doc esea ch i led Composing o Acous ic
Robo s - Ins an Syn hesis o Compu e -Con olled Acous ic Ins umen s Th ough
Li e Coding and AI. The COMACROB pe o mance aims o demons a e he
ad ancemen s on ins an composi ion in his ield ha ha e been ealized h ough
his pos doc esea ch. The esea ch was spli in wo pa s, whe e he i s pa
esul ed in a se ies o sho ins an composi ions o a Yamaha Diskla ie obo ic
piano, and he second pa esul ed in a single ins an composi ion o a MIDI-
con olled chu ch o gan. The COMACROB pe o mance mixes hese wo pa s,
by u ilizing he esul ing echnology ha was de eloped du ing his esea ch and
combining i wi h a human pe o me on a s anda d piano and elec onics.
1 Composing o acous ic obo s
Composing Fo Acous ic Robo s - Ins an Syn hesis o Compu e -Con olled Acous ic Ins umen s
Th ough Li e Coding and AI is he i le o he pos doc esea ch o Alexand os D ymoni is, unded by
he Cyp us Uni e si y o Technology. This esea ch commenced in July 2024 and ended on May 31s
2025. I was inspi ed by esea ch by D ymoni is (2023a) whe e AI was used o c ea e single ba s o
music and Bach’s cho als we e used as a p oo -o -concep . The so wa e used is he Li eLily sys em
o li e sco ing and li e sequencing h ough li e coding by D ymoni is (2023b). A Li eLily session
is shown in Figu e 1.
The COMACROB esea ch esul ed in wo conce s. The i s was a se ies o sho ins an compo-
si ions o a Yamaha Diskla ie obo ic piano, whe e eigh pianis s pe o med on i one a a ime,
alongside he ac ual obo ic piano which was con olled by a compu e . De ails on his pa o he
P oceedings o he 6 h Con e ence on AI Music C ea i i y (AIMC 2025),
B ussels, Belgium, Sep embe 10 h-12 h
esea ch can be ound in D ymoni is and Kou somichalis (2025). The second conce was an ins an
composi ion o a MIDI-con olled chu ch o gan ha was played bo h by a compu e and a human
pe o me . The aim o his esea ch was o in es iga e how AI can enhance li e sco ing h ough li e
coding, bu also how can he sigh - eading capaci y o a pe o me be augmen ed, by accompanying
hem wi h a compu e -con olled acous ic ins umen .
The AI was used as a way o c ea e music con en as , so ha a s eady pace could be e ained
h oughou he pe o mance. This was an a emp o add ess he issue o slow e olu ion in li e coding
pe o mances, as poin ed ou by Magnusson (2011) whe e he s a es ha " he e is li le un in wa ching
a s essed p og amme designing algo i hms o minu es be o e a simple sine oscilla o is applied
in he playback o a silly melody." By spli ing his esea ch in o wo pa s, I was able o es he
echnology used in he i s pa in a conce se ing and de e mine whe he i pe o med well enough
o add ess he ques ions o my esea ch. I u ned ou ha he AI echnology needed e inemen , and
in he second conce , he AI changed om code de i ed om he Classical Piano Compose
1
o he
No ocho d model by Shepa dson e al. (2022). The No ocho d model p o ided a esponsi e ool ha
could p o ide esh music con en as enough o keep he desi ed s eady pace. Being an al eady
ained model, using i libe a ed me om he cumbe some asks o assembling a aining da ase and
aining he AI model.
Ano he aspec o my esea ch was he e hical use o AI om a da a anspa ency and a c ea i e poin
o iew. In he i s conce I collabo a ed wi h ou compose s who p o ided o iginal music o be
used as a aining da ase . Thei names we e announced bo h a he beginning o he conce and in
D ymoni is and Kou somichalis (2025). The No ocho d model hough is ained on he Lakh MIDI
da ase
2
which consis s o 176,581 unique MIDI iles. Re e ing o all compose s o his da ase is
no possible, bu a link o c edi o hese pieces is a ailable
3
. Despi e using a ained AI model,
I used o iginal composi ions by Thanos Polymeneas Lion i is and mysel o p omp i du ing he
pe o mance, whe e again he names o he compose s we e men ioned a he beginning o he conce .
This way o using he No ocho d model is called in-con ex lea ning, whe e he model’s eac ions ge
close o i s p omp s as he pe o mance e ol es.
In addi ion o using o iginal composi ions, my pa icipa ion as a compose in his second conce
aimed a highligh ing he impo ance o he c ea i e ole o AI use s. In con as o many online
gene a i e AI se ices, whe e use s p omp AI models wi h a na u al language o c ea e music con en ,
bu ha e li le o no con ol in he ac ual c ea i e p ocess, I wan ed o s ess he impo ance o ha ing
an ac i e ole as a c ea o when using AI in music.
2 The AIMC conce
The COMACROB conce p esen ed a he AIMC is a combina ion o he wo conce s o my pos doc
esea ch, whe e he ins umen o he i s conce will be used wi h he AI echnology o he second
conce . In his pe o mance hough, he piano will no be a obo ic one, bu a classical one, played
by a human, wi h li e elec onics aking he ole o he compu e -con olled pa . To main ain he
app oach o an e hical use o AI in music, all he p omp s ha will be used du ing he pe o mance
will be pa o a composi ion o mine o piano and elec onics.
Alongside he use o AI in li e coding and li e sco ing, his pe o mance aims o p omo e he use
o li e coding combined wi h acous ic ins umen s, as his combina ion has no been ho oughly
explo ed. Mclean (2015) has explo ed such combina ions, bu li e sco ing is no p esen in his
esea ch. The COMACROB conce in oduces li e sco ing o his combina ion.
3 Me hodology and me hods
The me hodology o my esea ch is spli in o mic o-me hodology and mac o-me hodology. This
app oach was planned om he s a , as his pos doc esea ch was designed o ealize wo di e en
composi ions in wo conce s. A he beginning o his esea ch, he i s mic o-me hodoloy and he
mac o-me hodology we e laid ou , while he second mic o-me hodology was laid ou a e he i s
1h ps://gi hub.com/Skuldu /Classical-Piano-Compose /
2h ps://colin a el.com/p ojec s/lmd/
3h ps://colin a el.com/p ojec s/lmd/copy igh s. x
2
S age 1: Collec ion
o shee music in
MusicXML o ma
S age 2: C ea ion o
aining da ase s o AI
models
S age 3: T aining o
AI models and
es ing o hei
ou pu
Is ou pu
sa is ac o y?
S age 4: Rehea sals
wi h human
pe o me s and
compu e -con olled
ins umen s
A e
composi ions
sa is ac o y?
No Yes
YesNo
P esen a ion in
conce
Rewo k on
aining
da ase
Tes di e en
que ying
me hod
Figu e 2: Fi s i e a ion mic o-me hodology lowcha .
i e a ion o his esea ch had been ealized. Figu e 2 illus a es he mic o-me hodology o he i s
i e a ion, Figu e 3 illus a es he mic o-me hodology o he second i e a ion, while Figu e 4 illus a es
he mac o-me hodology.
The main di e ence be ween he wo mic o-me hodologies lies in he assembling o a aining
da ase o he AI and he ac ual aining p ocess, whe e in he second i e a ion no such p ocess ook
place. Ins ead o collec ing pieces and assembling a aining da ase , he i s s age o he second
mic o-me hodology consis ed o he collec ion o ewe pieces ha we e used only o que ying. The
i s mic o-me hodology could possibly oll back o wo s ages, depending on he pe o mance o he
AI model, while he second mic o-me hodology could ini ia e a loop wi hin a single s age, since no
assembling o a da ase o aining ook place.
The las s age o he wo mic o-me hodologies is he same. In his s age, ehea sals wi h he compu e -
con olled ins umen and pe o me s playing on i we e ealized. E en hough he AI ou pu had
al eady been es ed, his s age helped de e mine whe he he AI-gene a ed music con en was wi hin
he playing capaci y o he pe o me , in a sigh - eading se ing. Once he AI-ou pu was uned o i
he sigh - eading capabili ies o he pe o me s, I mo ed on o p esen his p ojec in a conce .
Acknowledgmen s and Disclosu e o Funding
The esea ch his pe o mance is based on has been unded by he POSTDOCTORAL p og am o he
Cyp us Uni e si y o Technology.
Re e ences
D ymoni is, A. (2023a). C ea ing music sco es li e h ough in e ac ing wi h a cha ac e -le el ex
gene a o .
3
S age 1: Collec ion
o shee music in
MusicXML o ma
S age 2: C ea ion o
Py hon sc ip o que y
AI model based on
shee music
S age 3: T ansla ion
o AI ou pu o he
[ANONYMOUS]
language
Is ou pu
sa is ac o y?
S age 4: Rehea sals
wi h human
pe o me s and
compu e -con olled
ins umen s
A e
composi ions
sa is ac o y?
No Yes
YesNo
P esen a ion in
conce
Tes di e en
que ying
me hod
Figu e 3: Second i e a ion mic o-me hodology lowcha .
D ymoni is, A. (2023b). Li elily: An exp essi e li e sequencing and li e sco ing sys em h ough li e
coding wi h he lilypond language. In O iz, M. and Ma quez-Bo bon, A., edi o s, P oceedings o
he In e na ional Con e ence on New In e aces o Musical Exp ession, pages 256–261, Mexico
Ci y, Mexico.
D ymoni is, A. and Kou somichalis, M. (2025). Composing o acous ic obo s - ins an syn hesis
o compu e -con olled acous ic ins umen s h ough li e coding and ai: Resea ch in p og ess.
Ch oma: Jou nal o he Aus alasian Compu e Music Associa ion, 40(1).
Magnusson, T. (2011). The IXI Lang: A Supe collide Pa asi e o Li e Coding. In P oceedings o
he In e na ional Compu e Music Con e ence, ICMC, pages 503–506, Hudde s ield, UK.
Mclean, A. (2015). Re lec ions on Li e Coding Collabo a ion. In P oceedings o 3 d con e ence on
Compu a ion, Communica ion, Aes he ics and X (xCoAx), pages 213–220, Glasgow, Sco land.
Shepa dson, V., A mi age, J., and Magnusson, T. (2022). No ocho d: a lexible p obabilis ic model
o embodied midi pe o mance. In P oceedings o he 3 d Con e ence on AI Music C ea i i y.
AIMC.
4
S age 1: E alua ion
o echnology and
aes he ics o i s
i e a ion
S age 2: E alua ion o
echnology and
aes he ics o second
i e a ion
Is
echnology
e icien ?
Tes di e en
echnology
A e
aes he ics
sa is ac o y?
No Yes
Use he same
echnology
Resea ch conclusion
Figu e 4: Mac o-me hodology lowcha .
5
