search

dansuh17 / segan-pytorch

SEGAN pytorch implementation https://arxiv.org/abs/1703.09452
GNU General Public License v3.0
106 stars 32 forks source link

the generated audio are nothing #30

Open hujinsen opened 5 years ago

hujinsen commented 5 years ago

the waveform of the denoised audio are zero. image

dansuh17 commented 4 years ago

Producing zero samples is most likely a symptom of discriminator failing. Have you checked the loss graph of discriminator?

SureArtificialIntelligence commented 3 years ago

Hi, did you solve this problem? I met the same problem at the beginning of training

SureArtificialIntelligence commented 3 years ago

The groundtruth is the raw audio, but the generated audio is range [-1, 1] due to Tanh() .......

The code does not work without modification