-
|
Hellow All, i could not find any information if Glow-TTS model requires vocoder model when i would like to training from scrach polish dataset model? if yes, vocoder will uses the same dataset like for Glow-TTS? Can i ask for advice? |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 5 replies
-
|
You don't necessarily have to train your own vocoder, you can use a pretrained one. If you're just running the Glow-TTS recipe it will only train the TTS model, the vocoder recipes are separate. |
Beta Was this translation helpful? Give feedback.
-
|
ok, so is there any advantage to have vocoder available during training model? And what is vocoder use for? |
Beta Was this translation helpful? Give feedback.
-
|
i have foud this https://github.com/jaywalnut310/glow-tts/blob/master/configs/base_blank.json which below statement: This result was not included in the paper. Lately, we found that two modifications help to improve the synthesis quality of Glow-TTS.; 1) moving to a vocoder, HiFi-GAN to reduce noise, 2) putting a blank token between any two input tokens to improve pronunciation. Does anyone try it? |
Beta Was this translation helpful? Give feedback.
-
|
Many TTS models, including Glow, only convert from text to Mel spectrograms. The vocoder is needed to convert these into a speech signal then, so that you can actually listen to the outputs. Coqui TTS has implementations and pretrained models for different vocoders, including HiFi-GAN, so you can pick your favourite. You can also train with such a blank token by setting |
Beta Was this translation helpful? Give feedback.
-
|
sorry for too many basic questions but i'm starting with this topic and i try to understand what to do in order to get some good results... So i can train glow model for polish language with HiFi-GAN vocoder pre-trained for other language? |
Beta Was this translation helpful? Give feedback.
-
|
thank you for all answers... |
Beta Was this translation helpful? Give feedback.
-
|
can this https://github.com/idiap/coqui-ai-TTS/blob/88de5c4f9be56b8e73a9ec002a8e04c48f12e00a/recipes/bel-alex73/train_glowtts.py#L61 be used as is or should i modify it based on my RTX2060 GPU hardware limitation? any sugesstion? |
Beta Was this translation helpful? Give feedback.
You don't necessarily have to train your own vocoder, you can use a pretrained one. If you're just running the Glow-TTS recipe it will only train the TTS model, the vocoder recipes are separate.