Commit c8b4b748 authored by Phil Wang's avatar Phil Wang
Browse files

fix l2 norm loss for log mel spectrogram differences in multi spectral recon... 

fix l2 norm loss for log mel spectrogram differences in multi spectral recon loss, thanks to @zhvng again
parent cb442af8

audiolm_pytorch/soundstream.py

+1 −6
Original line number Diff line number Diff line
@@ -33,11 +33,6 @@ def default(val, d): 
def cast_tuple(t, l = 1):

    return ((t,) * l) if not isinstance(t, tuple) else t



# tensor helpers



def l2norm(t, dim = -1):

    return F.normalize(t, dim = dim)



# gan losses



def log(t, eps = 1e-20):
@@ -617,7 +612,7 @@ class SoundStream(nn.Module): 
                log_orig_mel, log_recon_mel = map(log, (orig_mel, recon_mel))



                l1_mel_loss = (orig_mel - recon_mel).abs().sum(dim = -2).mean()

                l2_log_mel_loss = alpha * l2norm(log_orig_mel - log_recon_mel, dim = -2).mean()

                l2_log_mel_loss = alpha * (log_orig_mel - log_recon_mel).norm(p = 2, dim = -2).mean()



                multi_spectral_recon_loss = multi_spectral_recon_loss + l1_mel_loss + l2_log_mel_loss

setup.py

+1 −1
Original line number Diff line number Diff line
@@ -3,7 +3,7 @@ from setuptools import setup, find_packages 
setup(

  name = 'audiolm-pytorch',

  packages = find_packages(exclude=[]),

  version = '0.11.11',

  version = '0.11.12',

  license='MIT',

  description = 'AudioLM - Language Modeling Approach to Audio Generation from Google Research - Pytorch',

  author = 'Phil Wang',