Repolode
Submit
wenet-e2e

wenet-e2e/wenet

Production-oriented end-to-end speech recognition toolkit in PyTorch, with a command-line tool, Python API, training recipes and deployment runtimes.

View on GitHubWebsite
Stars
5,256
Forks
1,193
Open issues
22
Last commit
12 days ago

What is it?

WeNet is an end-to-end speech recognition toolkit built around production use. It covers the full path from installing a pretrained model and transcribing audio with a `wenet` command or Python API, to training new models and deploying them through runtimes.

The package offers pretrained models for Chinese (such as paraformer, firered and wenetspeech) and English (whisper-large-v3 and whisper-large-v3-turbo). Models run on CPU by default, and GPU use requires passing the device explicitly. The project also includes runtime builds for x86 and documents support for Ascend NPU setups.

Training and deployment are handled in a repository that requires cloning, a conda environment with PyTorch and torchaudio, and optional CUDA or CANN installation. Language model integration uses WFST-based decoding, which the project builds through its runtime.

Key features

  • 01Command-line transcription with pretrained Chinese and English models.
  • 02Python API for loading a model and transcribing audio in a few lines.
  • 03Pretrained models including paraformer, firered, wenetspeech and Whisper large-v3 variants.
  • 04Runtime builds for x86 and other platforms for deployment.
  • 05Support for Ascend NPU via torch-npu.

Use cases

  • Transcribing Chinese or English audio files from the command line or a Python script.
  • Training custom end-to-end speech recognition models in PyTorch.
  • Deploying streaming or non-streaming recognition services with the provided runtimes.
  • Experimenting with language model integration through WFST decoding.

Quick start

  1. 1

    Install the Python package from GitHub

    pip install git+https://github.com/wenet-e2e/wenet.git
  2. 2

    Transcribe an audio file from the command line

    wenet -m paraformer audio.wav
  3. 3

    Or use the Python API to transcribe audio

    python -c "import wenet; model = wenet.load_model('paraformer'); print(model.transcribe('audio.wav').text)"

Things to know

  • The command-line tool runs on CPU by default, so pass --device cuda to use a GPU.
  • Installing torch from pip may pick a build that does not match your GPU driver, so GPU use may need a manual reinstall.
  • Training and deployment setup requires Conda, CUDA and several extra dependencies such as sox.

In collections

AI video & audio tools

Topics

#speech-recognition#asr#pytorch#end-to-end#conformer#whisper#deployment#automatic-speech-recognition#e2e-models#production-ready

FAQ

What is wenet?
Production-oriented end-to-end speech recognition toolkit in PyTorch, with a command-line tool, Python API, training recipes and deployment runtimes. WeNet is an end-to-end speech recognition toolkit built around production use. It covers the full path from installing a pretrained model and transcribing audio with a `wenet` command or Python API, to training new models and deploying them through runtimes.
Is wenet free to use? What license does it have?
wenet is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is wenet still maintained?
Actively maintained. The last commit to wenet was on Oct 26, 2026.
What can I use wenet for?
Transcribing Chinese or English audio files from the command line or a Python script. Training custom end-to-end speech recognition models in PyTorch. Deploying streaming or non-streaming recognition services with the provided runtimes. Experimenting with language model integration through WFST decoding.
How do I install wenet?
Install the Python package from GitHub: pip install git+https://github.com/wenet-e2e/wenet.git Transcribe an audio file from the command line: wenet -m paraformer audio.wav Or use the Python API to transcribe audio: python -c "import wenet; model = wenet.load_model('paraformer'); print(model.transcribe('audio.wav').text)"
Where can I get wenet?
The source code and installation instructions are on GitHub: https://github.com/wenet-e2e/wenet

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects