winddori2002

Thank you for answering! I understand and solve it.

Hi, We have uploaded some audio pairs in the git (./samples). We randomly selected pairs and tried to choose some pairs containing more noises. Thanks.

Hi! We have uploaded the code for custom enhancement. You can find details in './custom_enhance.py'. The code is only for inference, and if you want to evaluate, you can refer...

I will upload the weights soon. Thank you.

Hi, We have uploaded both MANNER (small) and MANNER (base) a while ago. The weight you tried is MANNER (base) maybe. You can check the updated release. Thank you

Hi, I think it's ok since the speaker encoder indirectly learns to extract speaker identity. I tried other features such as wav2vec2.0, but it was less effective than CPC features....