Sequence level kD in fastspeech

as-ideas / TransformerTTS

🤖💬 Transformer TTS: Implementation of a non-autoregressive Transformer based neural network for text to speech.

https://as-ideas.github.io/TransformerTTS/

Other

1.12k stars 225 forks source link

Sequence level kD in fastspeech #99

Open bkumardevan07 opened 3 years ago

bkumardevan07 commented 3 years ago

Hi,

Did you not use sequence level knowledge distillation for fastpeech training??

cfrancesco commented 3 years ago

I used to, but I could not notice a difference in the results so to simplify the structure and to make it closer to the pre-trained vocoders I removed it.