OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Funasr

FunASR is an open-source Python toolkit and command-line tool for training and running speech recognition and related audio-processing pipelines. It supports documented offline and streaming ASR workflows alongside voice activity detection, punctuation, speaker processing, timestamps, and additional audio models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Funasr — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use FunASR when you need a developer-oriented toolkit to run speech models or compose transcription with related audio-processing stages, rather than a ready-made transcription website.

Who it is for

  • Developers building speech-to-text features or audio-processing workflows.
  • Teams running model-based transcription locally or in a self-hosted environment.
  • Researchers and practitioners who need examples and guidance for training or registering models.

Documented uses

  • Transcribe an audio file through Python or the command line.
  • Process audio in chunks for streaming recognition.
  • Combine recognition with voice activity detection, punctuation or speaker processing.
  • Export CLI results as JSON or SRT, including documented batch-transcription workflows.
Before you choose

Requirements & limits

The README documents installation from PyPI or source and links to a Colab quickstart for browser-based trials. Source installation requires Python 3.8 or later, with PyTorch and torchaudio installed first. CPU and GPU setup instructions differ.

  • Choose the model by task and language; the README states that coverage is checkpoint-specific.
  • GPU use requires matching PyTorch and torchaudio wheels and checking that PyTorch can see the GPU; CPU is an available alternative.
  • Streaming examples require chunked input and cache handling, unlike the single-file transcription example.
  • The toolkit has an MIT license, while individual model licenses may differ; the README identifies some model entries as third-party.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install FunASR and transcribe an audio file

Set up the Python toolkit and run a documented model on an audio input.

Before you start

  • Python 3.8 or later for the documented source installation.
  • PyTorch and torchaudio installed before FunASR.
  • For the documented GPU example, compatible PyTorch and torchaudio wheels and a visible NVIDIA GPU; otherwise use CPU.

Steps

  1. Install PyTorch and torchaudio, then install FunASR with the command below.

    Documented command
    pip install funasr

  2. Import AutoModel and select a model, such as FunAudioLLM/Fun-ASR-Nano-2512; choose device="cuda" only if torch.cuda.is_available() is true, otherwise use device="cpu".

  3. Call model.generate with an audio file or the documented public sample URL, then read the returned text.

Expected outcome

A Python result containing recognized text for the chosen model and input.

Guide limitations

  • Language coverage depends on the selected checkpoint.
  • The documented Fun-ASR-Nano example uses CUDA; the README links to a separate CPU quickstart using native Transformers.
  • Model licenses are separate from the toolkit license.
Documentation 1 ↗
Source-based practical guide

Transcribe audio with the command-line interface

Use the documented CLI examples to produce transcript or subtitle output.

Before you start

  • FunASR installed and an audio file available to the command line.

Steps

  1. Run funasr audio.wav for a basic transcription.

  2. Choose --output-format json for JSON output or --output-format srt --output-dir ./subs for subtitle output.

  3. For a batch example, run funasr *.wav --output-format srt --output-dir ./output.

Expected outcome

The CLI processes the specified audio input or batch and uses the selected output format and directory.

Guide limitations

  • The documented commands do not specify which model or languages are appropriate for every audio file.
  • The speaker-label and timestamp example is a separate command using --spk --timestamps -f json.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
GitHub - modelscope/FunASR: Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving. · GitHub · Official homepage title · retained in its source language
How we review sourcesSuggest a correction