FunASR v1.3.3 — vLLM Engine + Agent Integration + Ecosystem Update

[LauraGPT]

What's New in FunASR v1.3.3

vLLM Inference Engine

Fun-ASR-Nano now supports vLLM-powered batch inference — up to 393x realtime speed. WebSocket streaming service included.

pip install -U funasr
funasr-server --device cuda

OpenAI-Compatible API

funasr-server exposes a standard /v1/audio/transcriptions endpoint. Works with LangChain, AutoGen, Dify, and any OpenAI client.

MCP Server for AI Agents

Claude, Cursor, and Windsurf can now use FunASR for speech-to-text via the MCP Server.

---

Ecosystem

Project	Update
Fun-ASR-Nano	v1.0.0 released, 31 languages, Colab notebook
SenseVoice	Speaker diarization added
FunClip	Fun-ASR-Nano and SenseVoice support
www.funasr.com	New: FunASR vs Whisper comparison

---

Try It Now

• Colab Quickstart
• Live Demo (real-time WebSocket ASR)
• HuggingFace Space

Share your project in Show and Tell!

[penchy-zju]

wow，nice release