MASOM: Musical Agent based on Self-Organizing Maps

Kıvanç Tatar, and Philippe Pasquier
Can we use Artificial Intelligence (AI) to create super-human musicians? Can we create an AI that listens to more music than a human could? Can we train AI on the compositions of dead composers and play with them on stage?

Musical agents are AI software making music. Musical Agent based on Self-Organizing Maps (MASOM) is a musical software agent for live performance. MASOM plays experimental music and free improvisation. It learns by listening to audio files such as recordings of performances or compositions. We can train a MASOM agent on a set of music that is so big that it would take more than one human life to listen. Similarly, we can train MASOM agents on dead composers and convert their fixed media piece to interactive musical agents.

MASOM also extracts higher level features such as eventfulness, pleasantness, and timbre to understand the musical form of what it hears. MASOM is limited to the style of what is has listened to and reacts in real-time to what he is hearing. The agent can listen to itself and other performers to decide what to play next. MASOM is designed by Kıvanç Tatar and Philippe Pasquier.

Publications:

->Tatar K., Pasquier P., Siu R. (Upcoming-2019) Audio-based Musical Artificial Intelligence and Audio-Reactive Visual Agents in Revive. Accepted to the International Computer Music Conference and New York City Electroacoustic Music Festival 2019 (ICMC-NYCEMF 2019).
->Tatar K., Pasquier P., & Siu R. (2018). REVIVE: An audio-visual performance with musical and visual Artificial Intelligence Agents. CHI’18, April 21–26, 2018, Montreal, QC, Canada ACM 978-1-4503-5621-3/18/04. ->Tatar, K. & Pasquier, P. (2017). MASOM: A Musical Agent Architecture based on Self-Organizing Maps, Affective Computing, and Variable Markov Models. In Proceedings of the 5th International Workshop on Musical Metacreation (MuMe 2017). Paper

The research and development of MASOM is still ongoing. Following is the documentation of MASOM‘s previous versions, and public presentations. 

MASOM v1.30

Comparison of Factor Oracle, Variable Markov Model Max Order Variation, and Variable Markov Model Prediction by Partial Matching-C for Musical Agents based on Self-Organizing Maps

Please find examples of music generated by MASOM v1.30 below. Several models have been trained using two corpora of electroacoustic music and experimental electronic music with repetition. If you are interested in further details, please stay in tune for the upcoming paper.
All examples are also available here.

Corpus 1 - Electroacoustic Music 


Factor Oracle

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
All Corpus
Variable Markov Model Max-Order Variation (3rd Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
Variable Markov Model Max-Order Variation (5th Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4

Variable Markov Model Prediction by Partial Matching (3rd Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
All Corpus
Variable Markov Model Prediction by Partial Matching (5th Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
All Corpus


Corpus 2 - Experimental Music with Repetition 


Factor Oracle

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
Subset 5

All Corpus


Variable Markov Model Max-Order Variation (3rd Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
Subset 5


Variable Markov Model Max-Order Variation (5th Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
Subset 5


Variable Markov Model Prediction by Partial Matching C Variation (3rd Order)

Subset 0
Subset 1
Subset 2
Subset 3
Subset 4
Subset 5

Post-MUTEK 2018 Excertps





CHI 2018 - Performance at the SAT Montreal


with MASOM-Factor-rhythm v1.22

Copyright
Kıvanç Tatar ©2018-2019


Music

 ︎ Bandcamp:->Tatar ->Çekiç
︎ Soundcloud