How do you install Whisper AI?

Install Whisper by setting up Python and ffmpeg, then running one pip command. The official package is OpenAI's open-source Whisper project, used from the command line or from Python. You do not download a separate Whisper installer. Once the package is in place, one command transcribes an audio file, and the first run downloads a model.

Install Python, ffmpeg and the Whisper package, in that order

Type these commands in a terminal. On a Mac, open Terminal. On Windows, open PowerShell or Command Prompt. Use the same window for every step, so that the Python you installed is the one pip uses, and close it afterwards if you installed a new tool.

1. Install Python 3.8 to 3.11. The README says the codebase is expected to be compatible with Python 3.8 to 3.11 and recent PyTorch versions ¹. Check your version with python3 --version. If your default Python is newer, install a 3.11 alongside it and use that one for Whisper.
2. Install ffmpeg. openai-whisper requires the ffmpeg command-line tool on your system, in addition to the Python package ². The README lists package managers for each system: brew install ffmpeg with Homebrew on a Mac, and choco install ffmpeg with Chocolatey on Windows. Open a new terminal afterwards so the command is found.
3. Install the package. Run pip install -U openai-whisper ³. If pip fails while installing tiktoken, the README says you may need Rust, because tiktoken may not have a prebuilt wheel for your platform ⁴.

Transcribe one file to test the install

Run whisper audio.mp3 --model turbo. The README uses this command, which selects the turbo model ⁵. The turbo model needs about 6 GB of video memory ⁶. If your graphics card has less, or you have no suitable GPU, use a smaller model such as --model small, which needs less memory again. Our guide to running Whisper locally gives the memory and speed of every model.

The first run with a model downloads its weights. openai-whisper keeps them in ~/.cache/whisper unless you choose another folder with --model_dir ⁷. Plan for disk space as well as memory, because the model files are large.

Update the package, or fix a failed install

• To get the newest code instead of the last release on PyPI, the README gives a force-reinstall command ⁸: pip install --upgrade --no-deps --force-reinstall git+https://github.com/openai/whisper.git. Use it when you need a fix that has not reached PyPI yet.
• If pip or Python reports a version problem, check the Python version first, using the range in step 1.
• If you use Windows, the steps differ in places. Our guide to installing Whisper on Windows covers them.

If you would rather not use a terminal, the guide to running Whisper locally lists desktop apps that wrap the same engines. Our overview of speech recognition covers the subject as a whole. Some apps and keyboards also use the Whisper name, so check which software you are installing. Our answer on whether there is a Whisper AI app explains the difference.