- Part of the success of the OpenAI's MuseNet was in the accidental quantization of time. This allowed MuseNet model to learn to generate silent intervals and uneven tempos.
- Quantizing pitches also shows interesting results as the NN learns how to manipulate pitches properly.
- Other music elements such as duration and velocity can and should be quantized too. Unfortunately, this will dramatically increase the dataset size.
- Tokenized Sparse Time Quantization
- Tokenized Sparse Pitches Quantization
- If possible and resultative, Full Tokenized Sparse Quantization
@inproceedings{lev2021quantummusic,
title = {Quantum Music},
author = {Aleksandr Lev},
booktitle = {GitHub},
year = {2021},
}