Repolode
Submit
Const-me

Const-me/Whisper

Windows C++ port of OpenAI's Whisper speech recognition that runs inference on any Direct3D 11 GPU via DirectCompute.

View on GitHub
Stars
10,706
Forks
981
Open issues
172
Last commit
2 months ago

What is it?

Whisper is a C++ Windows implementation of OpenAI's Whisper automatic speech recognition model, built on the whisper.cpp project and its GGML model files. It runs the model's compute on the GPU through DirectCompute compute shaders, so it does not depend on CUDA or PyTorch and works with any hardware GPU that supports Direct3D 11.0.

The project ships a WhisperDesktop GUI for transcribing audio files and capturing live microphone audio, along with a COM-style native DLL and a C# wrapper distributed on NuGet. Audio decoding goes through Media Foundation, and voice activity detection is available for microphone capture. Only 64-bit Windows is supported.

The author reports that the project is no longer updated since 2023 and that a commercial successor exists.

Key features

  • 01Vendor-agnostic GPU inference using DirectCompute compute shaders on Direct3D 11.0 hardware
  • 02Native C++ library with no runtime dependencies beyond OS components
  • 03Desktop GUI for file transcription and live microphone capture
  • 04Audio and video decoding through Media Foundation
  • 05Voice activity detection for microphone capture
  • 06C# wrapper on NuGet and PowerShell 5.1 scripting support

Use cases

  • Transcribing audio files on a Windows PC without installing Python or CUDA
  • Adding offline speech recognition to a Windows C# or C++ application
  • Live transcription of microphone input on machines with integrated or non-NVIDIA GPUs

Quick start

  1. 1

    Download WhisperDesktop.zip from the Releases section, unpack it, and run WhisperDesktop.exe

  2. 2

    On first launch, download a model such as ggml-medium.bin (1.42GB) when prompted

Things to know

  • Windows 64-bit only; no macOS or Linux support.
  • Requires a CPU with AVX1 and F16C support.
  • The author has not updated the project since 2023, and a commercial successor exists.
  • Automatic language detection is not implemented, and real-time capture has high latency (about 5-10 seconds).

In collections

AI video & audio toolsRun and train AI on GPUs

Topics

#speech-recognition#whisper#asr#directcompute#gpu#windows#transcription

FAQ

What is Whisper?
Windows C++ port of OpenAI's Whisper speech recognition that runs inference on any Direct3D 11 GPU via DirectCompute. Whisper is a C++ Windows implementation of OpenAI's Whisper automatic speech recognition model, built on the whisper.cpp project and its GGML model files. It runs the model's compute on the GPU through DirectCompute compute shaders, so it does not depend on CUDA or PyTorch and works with any hardware GPU that supports Direct3D 11.0.
Is Whisper free to use? What license does it have?
Whisper is open source under the MPL-2.0 license. Check the license terms before using it in commercial products.
Is Whisper still maintained?
Updated within 6 months. The last commit to Whisper was on Sep 22, 2026.
What can I use Whisper for?
Transcribing audio files on a Windows PC without installing Python or CUDA Adding offline speech recognition to a Windows C# or C++ application Live transcription of microphone input on machines with integrated or non-NVIDIA GPUs
How do I install Whisper?
Download WhisperDesktop.zip from the Releases section, unpack it, and run WhisperDesktop.exe On first launch, download a model such as ggml-medium.bin (1.42GB) when prompted
Where can I get Whisper?
The source code and installation instructions are on GitHub: https://github.com/Const-me/Whisper

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects