(
Industrial speech recognition toolkit for offline, streaming, and edge deployment.
ASR · VAD · punctuation · speaker pipelines · emotion and audio-event models · OpenAI-compatible serving
Quick Start ·
MCP.Pizza Chef: modelscope
This speech-recognition toolkit lets your AI assistant listen to an audio file and hand back a clean transcript, complete with speaker labels, timestamps, and punctuation already added. A lighter model can also flag emotional tone, like laughter, and runs fine on an ordinary computer without a graphics card, while the top-accuracy model needs one. It connects directly to Cursor, and setup means installing it yourself with Python's pip or Docker rather than clicking a single button — no account or paid key required.
Communication
Data
Notes
Use This MCP server To
Transcribe a meeting recording with each speaker separated Turn a voice memo into text I can search or edit Add automatic timestamps and punctuation to a transcript Detect emotional tone or reactions like laughter in a recording Transcribe audio in real time as it streams in Convert speech in Chinese, English, or Japanese into text
README
FunASR FAQ
Can I use this to transcribe a meeting or interview recording?
Yes — it turns an audio file into text and can also tell you who was speaking and when, so you end up with a transcript broken out by speaker.
Can I use this to figure out who said what in a group conversation?
Yes — one of its built-in pipelines splits a recording into segments and labels each one with a speaker number, which is useful for meetings or interviews with more than one person talking.
Do I need an account or a paid key to use this?
No account or subscription key is required. You do need to install the toolkit and download the speech-recognition model files yourself, since everything runs on your own computer or server.
Which AI assistants or tools can I use this with?
It ships with a direct connection for Cursor and works with Claude-based assistants that support the same connection style. It also runs a server that mimics OpenAI's chat format, so it can plug into other tools built for OpenAI, Gemini, or similar assistants too.
How difficult is it to set up?
Setup takes some technical comfort — you install it with Python's pip or with Docker, and running the best-accuracy model usually calls for a computer with a graphics card. There is also a lighter model built for ordinary computers.
Can I use this without a powerful computer or graphics card?
Yes — the lighter SenseVoiceSmall model is built to run on an everyday computer's processor and still handle five-language transcription, though the flagship, higher-accuracy model does need a graphics card.
Does it work with languages other than English?
Yes — depending on which model you choose, it can transcribe Chinese, English, Japanese, and Chinese dialects, a broader set of 31 languages, or five languages plus emotional tone.
Will my audio be sent to a cloud service?
No — everything runs on your own computer or server that you set up, so recordings stay local unless you choose to deploy it somewhere else yourself.
Can it add punctuation and timestamps automatically?
Yes — the transcripts it produces already come with punctuation, timestamps for each spoken segment, and speaker labels included.