Congratulations on the Fun-CosyVoice 3.5 Release!

[linyueqian]

Hi team,

We're excited to see the announcement of Fun-CosyVoice 3.5 and Fun-AudioGen-VD. The FreeStyle instruction-controlled generation is a fantastic step forward!

We noticed the current GitHub repo is still on CosyVoice 3. We'd love to know if there are plans to open-source the 3.5 model weights and code, and if so, is there a rough timeline?

For context, we've been working on integrating CosyVoice into the vllm-omni project. Our PR for CosyVoice 3 support is nearly ready to merge (vllm-project/vllm-omni#498). We'd be very interested in collaborating to bring 3.5 support to vllm-omni as well. Having early access to the checkpoint (if open-source is planned) would help us prepare the integration in advance and ensure a smooth experience for the community on day one.

Of course, we fully understand if there's no confirmed timeline yet. Just wanted to express our interest and see if there's any way we can coordinate. Happy to discuss further!

Thanks for the amazing work 🙏

cc: @hsliuustc0106