search

lmnt-com / diffwave

DiffWave is a fast, high-quality neural vocoder and waveform synthesizer.
Apache License 2.0
754 stars 111 forks source link

Will unconditional waveform synthesis be available? #13

Closed chou852ishare closed 2 years ago

chou852ishare commented 3 years ago

Hello. Will unconditional waveform synthesis be available? And when? Thanks.

sharvil commented 3 years ago

I don't have any concrete plans for unconditional synthesis at the moment. I'd be happy to accept a contribution to the repository, though.

rave974 commented 3 years ago

what means uncoditional waveform synthesis?

sharvil commented 3 years ago

It means the model will produce a waveform without any additional input (i.e. no text, no spectrogram, etc.).

rave974 commented 3 years ago

why it's needed? if no text and no spectrogram what kind of waveform it makes?

sharvil commented 3 years ago

I recommend reading the DiffWave paper for details on unconditional generation and established results.

Andrechang commented 2 years ago

I don't have any concrete plans for unconditional synthesis at the moment. I'd be happy to accept a contribution to the repository, though.

I added unconditional option in my fork https://github.com/Andrechang/diffwave and add gtzan dataset and update torchaudio load_wav -> load