GitHub - anubhavgupta/whisper-npu: Speech to text at cursor using NPU. (shortcut -> win+/)

2 min read Original article ↗

Brings speech-to-text to any app at cursor via global hotkey.

  • Converts spoken words to text and puts that text at your current keyboard's cursor location.
  • Runs completely locally.
  • Keeps your GPU and CPU free by using Intel's NPU for running Speech to text conversion.

How to install

  1. Download the latest release from here: https://github.com/anubhavgupta/whisper-npu/releases/download/V1.0.1/whisper-npu.zip
  2. Extract the zip
  3. Start the whisper-npu.exe. (this auto downloads and uses the whisper-small model for you, model can be changed later)

How to use

  1. Hitting win + / keys would start capturing your audio
  2. Speak something
  3. Pressing the same keys again to stop capturing your audio
  4. The app puts the converted/transcribed text to the current focused input.

Please check Troubleshooting section in case if you face any issue

Configuration

Edit config.json to configure your setup:

{
  "sourceModel": "anubhav200/openai-whisper-tiny-openvino-int4", // check list of supported models below
  "modelRepositoryPath": "./downloaded-whisper-npu-models", // directory where models would be downloaded and stored
  "restPort": 8000, // port on which transciption server would run.
  "temperature": "0", // 0 to 1, can be adjusted to improve accuracy
  "language": "en", // set this to your speaking lang, this again will help improve accuracy
  "ovmsDirectoryPath": "./external/ovms",
  "soxDirectoryPath": "./external/sox/bin",
  "task": "speech2text",
  "targetDevice": "NPU" // models supports NPU only at the moment
}

Supported Models for NPU:

Following models are support to run on NPU, I will add more models soon. Change the value of sourceModel in config file to one of the following:

  • anubhav200/openai-whisper-tiny-openvino-int4
  • anubhav200/openai-whisper-small-openvino-int4
  • anubhav200/openai-whisper-large-v3-turbo-openvino-int4 (takes very long time to start)

Manual Build from source

Requirements

  • Node.js (v24 or higher)
  • Windows (for Windows Forms clipboard integration)
  • Machine with Intel NPU

1. Install Dependencies

2. Generate Executable

This will build and generate a ./pack/whisper-npu.zip file, extract it and run whisper-npu.exe to use the application.

Development

Building

Development Mode

Troubleshooting

  1. Make sure you have installed the latest Intel's NPU drivers by following the guide here: https://www.intel.com/content/www/us/en/support/articles/000099083/processors/intel-core-ultra-processors.html

License

MPL 2.0

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.