3 ms·

The first demo on that page was trained from a 32kbps crappy sound quality clip of winston churchill...? garbage in, garbage out?

• stavros 3y ago
  Must have been, it sounds very much like the quality of the "we shall fight on the beaches" speech. A bit unfortunate choice for a demo, sadly.
  • jpcl 3y ago
    Good point, thanks. And I was thinking it will show that the model can really synthesize very varied samples... ;)
    • stavros 3y ago
      It does, but maybe put it last!
      • jpcl 3y ago
        Did exactly that, thanks for spotting that. :) https://github.com/collabora/WhisperSpeech/commit/398b889060eb3ada5c4e4524dd5c2eb5a153515f https://github.com/collabora/WhisperSpeech/commit/398b889060...