Training higher quality audio for ~5 seconds

chrisdonahue / wavegan

WaveGAN: Learn to synthesize raw audio with generative adversarial networks

MIT License

1.33k stars 280 forks source link

Training higher quality audio for ~5 seconds #113

Open mattgallivan opened 2 years ago

mattgallivan commented 2 years ago

Is it possible to train 48 kHz audio for longer? I'm currently training with 65536 samples but this is not quite two seconds of resultant audio. I know I could add some convolutions to increase the size -- is there another way that might not drastically increase the parameters?