Qualcomm NPU (QNN, HTP)

This section explains how to run sherpa-onnx on a Qualcomm NPU using QNN with the HTP backend.

You’ll need a device equipped with a Qualcomm SM8350 processor or a newer, more powerful one, such as the SM8450, SM8475, SM8550, SM8650, SM8750, SM8850, and so on.

• Download QNN SDK
  • QNN_SDK_ROOT
• Build sherpa-onnx for Qualcomm NPU
  • Shared libraries
  • Executable files
• Run executables on your phone with adb (using model.bin)
  • Download a QNN model
  • Copy files to your phone
    • Copy model files
    • Copy sherpa-onnx executable files
    • Copy sherpa-onnx library files
    • Copy QNN library files
  • Run it !
    • Check files
    • Set environment variable ADSP_LIBRARY_PATH
    • Run sherpa-onnx-offline
  • Congratulations
• Run executables on your phone with adb (using libmodel.so)
  • Download a QNN model
  • Copy files to your phone
    • Copy model files
    • Copy sherpa-onnx executable files
    • Copy sherpa-onnx library files
    • Copy QNN library files
  • Run it !
    • Check files
    • Set environment variable ADSP_LIBRARY_PATH
    • Run sherpa-onnx-offline
      • Log of the first run
      • Log of later runs
  • Congratulations
• Build Android examples
  • Pre-built APKs
  • How to build Android examples
    • 1. Build shared libraries
    • 2. Copy QNN libs
    • 3. Download model files
    • 4. Change the code to use our selected model
    • 5. Build the demo
• Models for QNN
  • sherpa-onnx-qnn-SM8850-binary-5-seconds-zipformer-ctc-zh-2025-12-22-int8 (Chinese)
  • sherpa-onnx-qnn-SM8850-binary-5-seconds-zipformer-ctc-zh-2025-07-03-int8 (Chinese)
  • sherpa-onnx-qnn-SM8850-binary-5-seconds-sense-voice-zh-en-ja-ko-yue-2024-07-17-int8 (Chinese, English, Japanese, Korean, Cantonese, 中英日韩粤语)