Example: Serve FunASR/SenseVoice for speech recognition

[LauraGPT]

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 lightweight inference server framework!

Would you consider adding a FunASR/SenseVoice example for speech recognition serving?

Why?

• Most popular open-source ASR — FunASR 16K+ stars, SenseVoice 8K+
• Simple Python API — Perfect fit for LitServe
• Ultra-fast — SenseVoice-Small: ~70ms/10s, ideal for low-latency serving
• High demand — Many users want to deploy ASR as API

Example LitServe server

import litserve as ls
from funasr import AutoModel

class SpeechRecognitionAPI(ls.LitAPI):
    def setup(self, device):
        self.model = AutoModel(model="iic/SenseVoiceSmall", device=device)
    
    def predict(self, audio_data):
        result = self.model.generate(input=audio_data)
        return {"text": result[0]["text"]}

server = ls.LitServer(SpeechRecognitionAPI(), accelerator="auto")
server.run()

References

• FunASR: https://github.com/modelscope/FunASR (16K+ stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8K+ stars)

[LauraGPT]

Hi! FunASR/SenseVoice would be a great example for LitServe — it's a production-ready ASR with an OpenAI-compatible API. Happy to help with the integration! 🙏