First of all, thank you very much for contributing the code.
I used this project to train a very good Mandarin model.
But there is a problem. Every time I use short text to synthesize audio, the end of the audio will have a long silence.
When I am not using WaveRnn, I can use do_trim_silence=True
to valid vad (in the utils/syntheis.py
Line 81).
But when I use Wavennn as a vocoder, this method doesn’t work.
So Is there any way to stop the extra synthesis?
nmstoker
(Neil Stoker)
October 25, 2019, 3:48pm
2
Welcome to the forum.
It looks like you dual posted this verbatim, but did so ~4hrs after @erogol replied to your post in the originally posted issue! Best to stick to one location
opened 06:57AM - 25 Oct 19 UTC
closed 11:12AM - 28 Oct 19 UTC
First of all, thank you very much for contributing the code.
I used this projec… t to train a very good Mandarin model.
But there is a problem. Every time I use short text to synthesize audio, the end of the audio will have a long silence.
When I am not using WaveRnn, I can use `do_trim_silence=True` to valid vad (in the`utils/syntheis.py` Line 81).
But when I use WaveRnn as a vocoder, this method doesn't work. :(
So Is there any way to stop the extra synthesis?
[file_f9396704-dd1a-46d8-80f7-f75daf7544a0.wav.zip](https://github.com/mozilla/TTS/files/3770736/file_f9396704-dd1a-46d8-80f7-f75daf7544a0.wav.zip)
Sorry, I actually sent it at the same time, but because I just registered my account,It took time to review.orz,I saw his reply, thank you.
Next I will communicate with him on github.thx~:smiley:
nmstoker
(Neil Stoker)
October 26, 2019, 8:35am
5
Ah, I see! Then it is I who am sorry for my misunderstanding. Glad you’ve got an answer!