The content of the generated sound is not correct #358

New Issue

Mo-Jiaxuan · 2023-08-30T12:48:05Z

Mo-Jiaxuan commented

2023-08-30 12:48:05 +00:00

Is anyone else experiencing the same problem? I've fine-tuned the model using a Japanese dataset, and when I use the trained model to generate the audio, it doesn't come back with what I've provided!
Where might this be caused by a problem?
I can provide any details of the training process

Is anyone else experiencing the same problem? I've fine-tuned the model using a Japanese dataset, and when I use the trained model to generate the audio, it doesn't come back with what I've provided! Where might this be caused by a problem? I can provide any details of the training process

mrq commented

2023-08-30 17:44:02 +00:00

The issue I've ran into with naively using Japanese is that there's a problem with the way the default tokenizer will normalize Japanese text (it will convert kana/kanji the wrong way). I honestly don't remember which issue I mentioned this in, and if said issue contained my notes on it again, but it's imperative that you swap over to the provided Japanese tokenizer for training.

Sign in to join this conversation.