Introduction

sherpa is the deployment framework of the Next-gen Kaldi project.

sherpa supports deploying speech related pre-trained models on various platforms with various language bindings.

If you are interested in how to train your own model or fine tune a pre-trained model, please refer to icefall.

At present, sherpa has the following sub-projects:

• k2-fsa/sherpa

• k2-fsa/sherpa-onnx

• k2-fsa/sherpa-ncnn

The differences are compared below:

	k2-fsa/sherpa	k2-fsa/sherpa-onnx	k2-fsa/sherpa-ncnn
Installation difficulty	hard	easy	easy
NN lib	PyTorch	onnxruntime	ncnn
CPU Support	x86, x86_64	x86, x86_64, arm32, arm64	x86, x86_64, arm32, arm64, **RISC-V**
GPU Support	Yes (with CUDA for NVIDIA GPUs)	Yes	Yes (with Vulkan for ARM GPUs)
OS Support	Linux, Windows, macOS	Linux, Windows, macOS, iOS, Android	Linux, Windows, macOS, iOS, Android
Support batch_size > 1	Yes	Yes	No
Provided APIs	C++, Python	C, C++, Python, C#, Java, Kotlin, Swift, Go, JavaScript, Dart Pascal, Rust	C, C++, Python, C#, Kotlin, Swift, Go
Supported functions	streaming speech recognition, non-streaming speech recognition	streaming speech recognition, non-streaming speech recognition, text-to-speech, speaker diarization, speaker identification, speaker verification, spoken language identification, audio tagging, VAD, keyword spotting,	streaming speech recognition, VAD,

We also support Triton. Please see Triton.