Spaces:

ResembleAI
/

Chatterbox-Multilingual-TTS-V3

Running on Zero

App Files Files Community

Space inference broken/buggy

by KevinAHM - opened Jun 11

Discussion

KevinAHM

Jun 11

•

edited Jun 11

Generates noticeably poor TTS with some references compared to v2 - hallucinations and wrong prosody. Same input (default settings, English reference and text) works fine on the old multilingual model and space. One reference had it generate 30s of silence after 10s output.

tedi-resemble

Resemble AI org Jun 11

We're looking

Zihan428

Resemble AI org Jun 11

Could you send the reference clip that causes hallucinations? @KevinAHM

KevinAHM

Jun 11

Could you send the reference clip that causes hallucinations? @KevinAHM

Yes, please provide an email or other method of private contact.

Zihan428

Resemble AI org Jun 12

Thank you! zihan@resemble.ai @KevinAHM

Zihan428

Resemble AI org Jun 15

Hi @KevinAHM the issue is fixed! I sent the updated output through email. Thanks for the feedback!

Zihan428 changed discussion status to closed Jun 15

puregreen

Jul 22

Hi @KevinAHM the issue is fixed! I sent the updated output through email. Thanks for the feedback!

could you please clarify what the issue was and if we should change something in the code?

mpasila

Aug 5

Is it actually fixed? Because in comparison to V2 the Finnish generation is much worse and less natural than V2. V2 was pretty natural, it pronounced most stuff correctly, had a bit of a weird accent sometimes but this new one just sounds like an English speaker trying to speak Finnish even when using a Finnish voice sample.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment