To run the Whisper CLI on Windows, you have two main options: using the officialPython packagewithpipor using the standalone executables provided by the C++ port,whisper.cpp. The Python method offers direct installation, whilewhisper.cppprovides better performance on CPU hardware.
Option 1: Using the Official Python Package (Recommended for most users)
This method involves using Python's package manager (pip) to install the official OpenAI Whisper library, which includes a command-line interface.
Prerequisites:
Python: Install Python 3.8 or later from the official website. Ensure you check the box that says "Add Python to PATH" during installation.
FFmpeg: This tool is required for handling various audio formats.
The easiest way to install it on Windows is using the package manager Chocolatey.
Open PowerShell as an administrator and run: choco install ffmpeg.
Alternatively, you can manually download FFmpeg and add it to your system's environment variables (PATH).
PyTorch (Optional for GPU acceleration): If you have an NVIDIA GPU, install CUDA-compatible PyTorch for much faster transcription. Follow instructions on the PyTorch website. For CPU-only, the pip install command will install the necessary CPU version automatically.
Installation Steps:
Open your Command Prompt or PowerShell.
Install the Whisper package using pip:
bash
pip install -U openai-whisper
To ensure you have the latest version directly from the source, you can use this command instead:
Usage Example: Transcribe an audio file named audio.mp3 using the small model:
bash
whisper audio.mp3 --model small
Option 2: Using whisper.cpp (For CPU performance and standalone use)
whisper.cpp is a C++ port that runs very efficiently on a CPU, often faster than the Python version. Standalone CLI executables are available, eliminating the need to install Python or manage dependencies manually.
Download: Go to the whisper.cpp releases page on GitHub and download a pre-built Windows binary (e.g., whisper-bin-x64.zip).
Extract: Extract the contents of the ZIP file to a folder (e.g., C:\ai\whisper.cpp).
Install C++ Redistributable: Ensure you have the latest Microsoft Visual C++ Redistributable installed, as it is a necessary dependency.
Download a Model: Download a pre-trained model file in the ggml format from the Hugging Face repository (e.g., ggml-base.en.bin) and place it in your whisper.cpp folder.
Run: Open a Command Prompt in the extracted folder and run the executable with your model and audio file.