Discrete content encoder example about soft-vc

[bshall]

Hi @yd11111, to use the discrete content encoder you have to unsqueeze the units

with torch.inference_mode():
    units = hubert.units(source).unsqueeze(0)
    mel = acoustic.generate(units).transpose(1, 2)
    target = hifigan(mel)

I'll add this to the .units() method to make it cleaner but you can manually add it in for now.

[yd11111]

Hi @bshall , I have worked it out and generated some interesting voice converted waves, many thanks for your interesting work.