-
|
This is an incredible project. Well done to the creator! I have 2 questions regarding the program.
|
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 2 replies
-
|
Hey, thanks! Ack on the long sentence problem. This is mostly caused by the fact that my training set-up could not accommodate training against long speech segments, so the models just get weaker in these cases. Still, it should do a pretty good job up to 15 seconds, which is more than enough time to speak even fairly long sentences. For extra-long sentences, I recommend breaking on commas, semicolons or dashes. You can precompile voice latents using |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the reply. The long sentences seem to be a problem when there is text in brackets mid sentence. I'm messing with the text brake up code atm so it includes that. |
Beta Was this translation helpful? Give feedback.
-
|
Ahhh I understand now. Yeah that's probably going to get problematic with too much prompting text. Sorry. :( |
Beta Was this translation helpful? Give feedback.
-
|
Here is the detailed guide on How to Generate Longer Speech with Tortoise TTS on Colab |
Beta Was this translation helpful? Give feedback.
Hey, thanks!
Ack on the long sentence problem. This is mostly caused by the fact that my training set-up could not accommodate training against long speech segments, so the models just get weaker in these cases. Still, it should do a pretty good job up to 15 seconds, which is more than enough time to speak even fairly long sentences. For extra-long sentences, I recommend breaking on commas, semicolons or dashes.
You can precompile voice latents using
get_conditioning_latents.py: https://github.com/neonbjb/tortoise-tts#generating-conditioning-latents-from-voices. Expect to shave off about half a second by doing this.