Overview of Whisper
Whisper is a speech recognition model released by OpenAI as code and model weights on GitHub. It performs multilingual speech recognition, speech translation, language identification and voice activity detection, and can be run from a command line or from Python.
The repository offers six model sizes, tiny through large plus turbo, ranging from 39M to 1550M parameters. Turbo is described as an optimized version of large-v3 with faster transcription and minimal accuracy loss, though it is not trained for translation.
Because the weights are MIT-licensed, developers and researchers can run it locally rather than through a hosted service. Setup requires ffmpeg and a Python install.