OpenAI's Whisper turns audio into accurate, multilingual transcripts entirely on your own PC no internet needed once it's set up, no per-minute fees.
Get it from GitHubOpens OpenAI's official Whisper repository · Windows, macOS, Linux
Maker
OpenAI
Cost
Free
Runs
Fully offline
Languages
~99
The original OpenAI package. Command-line only, installed with pip — flexible but needs Python set up first.
GitHub repo →A GPU-accelerated Windows port with a proper desktop interface — drag in a file, get a transcript, no command line.
GitHub repo →Install Python 3.10+ and FFmpeg
Get Python from python.org, then download FFmpeg and add its bin folder to your system PATH.
Open Command Prompt and install Whisper
Run pip install openai-whisper — this pulls the package straight from PyPI.
(Optional) Set up CUDA for GPU speed
An NVIDIA GPU with CUDA speeds transcription up significantly; Whisper falls back to CPU automatically if you skip this.
Transcribe a file
Run whisper yourfile.mp3 --model small in the same folder as your audio file.
| tiny / base | Fastest, runs fine on CPU, lower accuracy — good for quick drafts |
| small | Solid balance of speed and accuracy for most everyday use |
| medium | Noticeably more accurate, wants a decent GPU to stay fast |
| large / turbo | Best accuracy, slow on CPU-only machines, ~10 GB+ VRAM ideal |