New chatterbox with multi-language support
First, thanks for creating this model.
Do you plan to build a new version with the actual chatterbox multi-language base?
It already supports german, but maybe it could get even better with your dataset.
Thanks 🤗 also for the note on the new model.
I tried it and it sounds really good in german. Is there anything else missing?
The tokenizer seems to have [laughter], ... tokens but they are not working. Anything else?
I have to say that the quality of the multi-language model doesn’t come close to your finetune here. Certain words in German are pronounced somewhat “oddly,” and I also hear noticeably more artifacts. All in all, your finetune here satisfies me much more.
hey Sebastian,
What do you say about this model? https://huggingface.co/openbmb/VoxCPM-0.5B
For now its Englisch and Chinese only. Are you in the position to teach this model German?
hey Sebastian,
What do you say about this model? https://huggingface.co/openbmb/VoxCPM-0.5B
For now its Englisch and Chinese only. Are you in the position to teach this model German?
Thank you for mentioning it. I gave it a try and it's actually really good, better than Chatterbox in my first tests.
Cool! Someone on Reddit explained to me following:
„ No tokenizer and small/medium size means it should be finetunable, hoping unsloth guys have some love to make this fast and doable.“
Hey,
not planning to finetune it as there is no finetuning code yet. DiT is a bit harder to implement compared to the easy LLM loss function. If there is ft code let me know😀
But I am trying out the multilingual chatterbox model, maybe an update will follow later.