search

coqui-ai / TTS

🐸💬 - a deep learning toolkit for Text-to-Speech, battle-tested in research and production
http://coqui.ai
Mozilla Public License 2.0
35.8k stars 4.38k forks source link

[Bug] tts_models/en/vctk/fast_pitch does not generate speech properly #981

Closed wangtiance closed 3 years ago

wangtiance commented 3 years ago

Describe the bug I installed TTS 0.4.1 and ran tts --model_name tts_models/en/vctk/fast_pitch --text "hello world." --speaker_idx VCTK_p225 It uses vocoder_models/en/vctk/hifigan_v2 by default, but the wav file generated contains pure noise. I tried a bunch of different speaker_idx and the results are the same. I also tried other vocoders like vocoder_models/universal/libri-tts/fullband-melgan, and the output is barely intelligible.

Other models like tts_models/en/ljspeech/tacotron2-DDC worked fine for me, but I wanted to try multi speaker models.

Environment (please complete the following information):

tts_output.zip

skol101 commented 2 years ago

Yep, that's a known issue. https://github.com/coqui-ai/TTS/issues/907