Run OpenAI Whisper on Windows with GPU in 5 Minutes!

Summary

The presenter demonstrates a native Windows graphical interface for running OpenAI Whisper with GPU acceleration. The app is a Windows port of whisper.cpp, a C++ implementation of Whisper, and is distributed under the MPL 2.0 license. The presenter recommends reviewing the license before commercial use.

Setup guide

1. Download the Whisper Desktop ZIP from the project’s GitHub Releases page and launch the application.
2. Download a GGML medium model from Hugging Face. The presenter reports a file size of about 1.4–1.5 GB.
3. Use the app’s browse option to select the model, then choose a GPU. The presenter selects an NVIDIA G4 device. The app loads the model before use.

The presenter notes that model selection and downloading could be clearer: the app links to an external Hugging Face page, and some setup options are not well explained.

Features and test results

• Microphone capture: The app can listen to a selected input device and transcribe speech, displaying results with some delay. Transcriptions can be saved to a text file, with options to append text and include timestamps.
• Audio and video transcription: The presenter tests an MKV video file and confirms that the app can transcribe its audio.
• Speed and quality: A 3-minute-32-second video takes about 26 seconds to process, or roughly 7.9× faster than real time. The presenter considers the transcript generally good but notes at least one recognition error.

Source

The video’s presenter is from Open Integrator and demonstrates the Whisper Desktop Windows interface and the associated Whisper.cpp project.