Feature: Add FunASR/SenseVoice speech models

[!NOTE] License and capability clarification (2026-07-14): FunASR is a toolkit, not a single checkpoint. The FunASR and SenseVoice repository source code is MIT; model weights follow each model card. SenseVoiceSmall supports Chinese, Cantonese, English, Japanese, and Korean, and its weights use the linked FunASR Model Open Source License Agreement. Fun-ASR-Nano-2512 is Apache-2.0. Language coverage, punctuation, and performance depend on the selected model and runtime configuration.

Hi! Great framework for serving open-source LLMs!

Would you consider adding speech models like FunASR / SenseVoice?

Why?

OpenLLM focuses on LLMs, but speech LMs are becoming a major category:

• Fun-ASR-Nano — Audio encoder + Qwen2.5-0.5B LLM, end-to-end speech understanding
• SenseVoice — Multi-task speech model (ASR + emotion + audio events), ~70ms/10s
• Paraformer — Non-autoregressive ASR, production-grade

These are some of the most popular open-source speech models (17K+ and 8K+ stars) and could expand OpenLLM into the speech modality.

References

• FunASR: https://github.com/modelscope/FunASR (17K+ stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8K+ stars)
• Fun-ASR-Nano: https://github.com/FunAudioLLM/Fun-ASR (1.2K stars)