modelscope--funasr/examples/dify

FunASR + Dify Integration Guide

Use FunASR as a local speech-to-text backend for Dify via the OpenAI-compatible API.

Why FunASR with Dify?

• Privacy: All audio stays local, no cloud API calls
• Speed: 170x realtime (vs ~1x for cloud APIs)
• Free: No per-minute charges, no API keys needed
• Features: Speaker diarization, emotion detection, 50+ languages

Setup

1. Start FunASR Server

pip install torch torchaudio
pip install funasr vllm fastapi uvicorn python-multipart
funasr-server --device cuda --port 8000

2. Configure in Dify

In Dify Settings -> Model Provider:

1. Select OpenAI-API-Compatible provider
2. Set:
   • API Base URL: http://your-server:8000/v1
   • API Key: unused (any value works)
   • Model Name: fun-asr-nano

3. Enable Speech-to-Text

In your Dify app settings:

1. Go to Features -> Speech to Text
2. Enable it
3. Select the OpenAI-compatible provider you configured

Supported Models

Model Name	Best For	Speed
fun-asr-nano	31 languages + Chinese dialects	340x realtime (vLLM)
sensevoice	Ultra-fast, 5 languages	170x realtime
paraformer	Chinese (classic, stable)	120x realtime

Links

• FunASR GitHub
• Dify GitHub
• FunASR Website