can i use the multiband-melgan vocoder trained on ljspeech with a tacotron trained on a different language dataset?
Technically this works as long as your Taco2 model output is compatible with the vocoder model input format. You should be aware that you will get the LJspeech voice speaking with a strong accent of your Taco language.
1 Like
there are universal vocoder models on the release models page. So better use one of them.
1 Like
while the taotron2 model is being trained test audio files gets generated…in this case which vocoder is being used to convert those mel spectrogram to audios?
The „built-in“ Griffin-Lim vocoder.
1 Like