search

Olaf2K / Neural_TTS

This is a custom python script using Mozilla neural TTS that takes a batch of input files as text, and turns them into one long .wav file.
1 stars 1 forks source link

speedy-speech-wn #1

Open rejuce opened 3 years ago

rejuce commented 3 years ago

i would like to change tacotron to speedy-speech-wn tts model in the script. any chance you can point me on where i need to change that ? ( i did not find the string directly in the script)

the model is listed when i type tts --list_models and i find it better sounding than tacotron2

Olaf2K commented 3 years ago

Not too sure, probably best to ask this on the main git: https://github.com/mozilla/TTS

Please do let me know if you figured out if they support it and I can have a look if my loop works for that too.

rejuce commented 3 years ago

thank you for your reply. i think your idea is great of splitting it into sentences and querry sentence by sentence and merge in the end.

i try a differnt approach now, i forked your respository and i am replaceing the tts part with a http call to the tts webserver/docker container. that way the text splitting / wav mergeing part becomes decoupled from the tts. it is already working but occasionally i get the last word of an sentence repeated idefinitely, ahve to lookt into it closer... (i am not profecient in python unfortunately...)

my ultimate goal is to convert whole ebooks with it. if that works i might also build a small Qt frontend for it

Olaf2K commented 3 years ago

That sounds very cool, please get in touch if you need any help with the python part. In all honesty, when I build this I just started with python, so the script is far from being elegant.