wenet

Why we included this project

WeNet is built around the idea that ASR should survive contact with real deployments, not just demos. You train Conformer and Transformer models in PyTorch, then push the same pipeline into a prebuilt runtime that runs on Linux servers, Android, iOS, Raspberry Pi, and NPUs. For Chinese speech it ships with useful pretrained checkpoints like Paraformer and WenetSpeech, while Whisper backends cover English through the same interface. The pip package gives you a command line and Python API, so you can transcribe an audio file before touching the heavier training setup.