PerformanceRNN

Preprocess

Event-based representation: note-on, note-off, time-shift, velocity에 대한 event를 one-hot vector로 표현하여 sequentially 나열한다.

 

Model

PerformanceRNN

- Autoregressive LSTM

- Input as 413-dimensional one-hot vectors

- Teacher forcing for training

- Sampling output from beam search

 

이외에도 music generation에 대한 전반적인 개요를 설명해주는 좋은 논문.

 

References

[1] S. Oore et al., This Time with Feeling: Learning Expressive Musical Performance, Neural Computing and Applications 2018