Is there a maximum limit of characters for a single inference?

[AznamirWoW]

Trying to run TTS on <200 characters seems fine, but on a larger test (~1600 characters) the result is a random mix of phrases in a random order.

[sekstini]

Yeah, the model was trained on clips of at most 30 seconds, so long prompts can be out of distribution. The way we handle this on the website is by chunking the text and generating audio for the chunks one at a time.

[6630507]

In the gradio interface, I get a loud and constant buzz like a sawtooth wave for anything after :30 sec.

Also happens after :30 sec using the same text in the sample python code on the project page.

[FurkanGozukara]

Yeah, the model was trained on clips of at most 30 seconds, so long prompts can be out of distribution. The way we handle this on the website is by chunking the text and generating audio for the chunks one at a time.

can you add this auto chunking to the gradio please ty

[spaghettiSystems]

we may add it to the public inference code at some point but it's not a priority. you're invited to contribute it however

[InconsolableCellist]

I have a PR for this here