SenseVoice

This section describes how to use models from https://github.com/FunAudioLLM/SenseVoice.

A single model from SenseVoice supports the following languages

• Chinese (Mandarin, 普通话)

• Cantonese (粤语, 广东话)

• English

• Japanese

• Korean

In the following, we describe how to download pre-trained SenseVoice models and use them in sherpa-ncnn.

• Export SenseVoice to sherpa-ncnn
  • The code
• Pre-trained Models
  • sherpa-ncnn-sense-voice-zh-en-ja-ko-yue-2024-07-17 (Chinese, English, Japanese, Korean, Cantonese, 中英日韩粤语)
    • Download
    • Decode a file
      • Without inverse text normalization
      • With inverse text normalization
    • Real-time Speech recognition from a microphone
    • Speed test on RK3588 CPU
      • Cortex A55
      • Cortex A76
  • sherpa-ncnn-sense-voice-zh-en-ja-ko-yue-2025-09-09 (Chinese, English, Japanese, Korean, Cantonese, 中英日韩粤语)
    • Download
    • yue-0.wav
    • yue-1.wav
    • yue-2.wav
    • yue-3.wav
    • yue-4.wav
    • yue-5.wav
    • yue-6.wav
    • yue-7.wav
    • yue-8.wav
    • yue-9.wav
    • yue-10.wav
    • yue-11.wav
    • yue-12.wav
    • yue-13.wav
    • yue-14.wav
    • yue-15.wav
    • yue-16.wav
    • yue-17.wav