numbers game by pulu

## numbers game ##

a sequence of neural network explorations
plus a few other pieces

we are only beginning to find out the sonic possibilities of neural networks; on this album i explore some techniques for sound-morphing, weird sampling and multidimensional granular synthesis.

tracks 1-10 created with deep learning tools:

1 - GANSpaceSynth
2 - GANSpaceSynth, Hallu
3 - Open NSynth Super, SuperCollider, upright piano
4 - GANSpaceSynth, Hallu
5 - SampleRNN
6 - GANSpaceSynth, AI-terity 2.0 debugging
7 - GANSpaceSynth
8 - SampleRNN
9 - SampleRNN
10 - GANSpaceSynth, AI-terity 2.0 development version

## addendum (2024) ##

when this album was created in 2021, i naively did not foresee the disappointing direction that mainstream generative "AI" would take in the next few years. unlike today's models trained on vast amounts of data non-consensually scraped from all over the web, the models used here are small and mostly trained on my own music.

even so, our research left me doubtful about whether the sonic output of these models is worth the energy used to train and run them. later, seeing generative "AI" become a tool for tech companies to extract revenue from the creative work of artists, while also aiming to supplant us, soured me on the entire field.

all of this has inspired me to explore music-technological approaches for doing more with less. i would not make this album again, but i am leaving it up for posterity. i think the non-deep-learning bonus tracks here were always better anyway :)

models on tracks 1, 2 and 7-10 trained on my music
model on track 3 trained on the open-source NSynth dataset
model on track 4 trained on my personal collection of drum samples
model on track 5 trained on the music of KOM-Teatteri

miranda. computer sound worker

pulusound.fi