VFWS : VoiceFilter Without Speaker

Using Voicefilter to noise surppression only without targeted speaker embedding.

Conclusion

Not Good

Result

log magnitude, L1 loss, oneCycleLR, step 40k, epoch 3

Smudging harmonic components.

Based on https://arxiv.org/abs/1810.04826
and https://github.com/mindslab-ai/voicefilter