During speech recognition, it does not need to access the Internet. Everyting is processed locally on your device.

We support using onnx with onnxruntime to replace PyTorch for neural network computation. The code is put in a separate repository sherpa-onnx .

sherpa-onnx is self-contained and everything can be compiled from source.

Please refer to https://k2-fsa.github.io/icefall/model-export/export-onnx.html for how to export models to onnx format.

In the following, we describe how to build sherpa-onnx for Linux, macOS, Windows, embedded systems, Android, and iOS.

Also, we show how to use it for speech recognition with pre-trained models.