add more speech isolation models like MossFormer2_SE_48K, MossFormerGAN_SE_16K, FRCRN_SE_16K, Noisy, DEMUCS-for-Speech-Enhancement

[insinfo]

add more speech isolation models like

• MossFormer2_SE_48K,
• MossFormerGAN_SE_16K,
• FRCRN_SE_16K,
• Noisy,
• DEMUCS-for-Speech-Enhancement,
• MP-SENet (Magnitude-Phase Speech Enhancement Network),
• DCCRN (Deep Complex Convolution Recurrent Network),
• SEGAN (Speech Enhancement GAN),
• MetricGAN+

https://github.com/JaeBinCHA7/DEMUCS-for-Speech-Enhancement
https://github.com/modelscope/ClearerVoice-Studio/tree/main/clearvoice

[RyanMetcalfeInt8]

Thanks for the recommendation. Could you (and others) help evaluate them and let me know which ones produce the highest quality output? This kind of evaluation is very time consuming for us.

[MarcoRavich]

Well, DEMUCS-for-Speech-Enhancement is pretty dated, but ClearerVoice-Studio isn't (and thay do have models on Huggingface) so I do believe is the way to go.

Anyway CV does perform multiple speech-oriented enhancements, not only separation...

As always I suggest to check my resources' collection (under HyMPS \ AUDIO \ AI-based \ Voicing) to identify at a glance which projects are up to date.

[RyanMetcalfeInt8]

@MarcoRavich, as always, thanks for your insights! ClearerVoice-Studio looks like a fantastic source to pick up some high quality, pre-trained models. Been having a harder time this year squeezing in time to implement new features, but am very interested to dive into it.