Repolode
Submit
openai

openai/whisper

General-purpose speech recognition model that transcribes, translates and identifies the language of audio across many languages.

View on GitHub
Stars
110,909
Forks
13,448
Open issues
175
Last commit
2 months ago

What is it?

Whisper is an open-source speech recognition model from OpenAI, trained on a large and diverse set of audio. It handles multilingual speech recognition, speech-to-English translation, and spoken language identification within a single Transformer sequence-to-sequence model, so it can replace several stages of a traditional speech processing pipeline.

It is available as a Python package (installable with pip) and as a `whisper` command-line tool that needs ffmpeg for audio decoding. Six model sizes are provided, from `tiny` to `large`, plus a `turbo` model optimized for faster transcription; English-only `.en` variants exist for the smaller sizes. Larger models need more VRAM, from about 1 GB for tiny up to about 10 GB for large.

Beyond the high-level transcribe call, the Python API exposes lower-level functions for language detection, log-Mel spectrogram creation and decoding options. The `turbo` model is not trained for translation, so translation tasks should use the multilingual medium or large models.

Key features

  • 01Multilingual speech recognition in one model.
  • 02Speech-to-English translation with multilingual models.
  • 03Automatic spoken language identification.
  • 04Six model sizes trading accuracy against speed and memory.
  • 05Command-line tool and Python API for transcription.

Use cases

  • Transcribing meeting recordings, interviews or podcasts into text.
  • Adding subtitles or captions to audio and video content.
  • Translating non-English speech into English text.
  • Building voice input features into applications or local tools.

Quick start

  1. 1

    Install ffmpeg using your system package manager (e.g. Homebrew on macOS)

    brew install ffmpeg
  2. 2

    Install Whisper from PyPI

    pip install -U openai-whisper
  3. 3

    Transcribe an audio file from the command line

    whisper audio.flac audio.mp3 audio.wav --model turbo

Things to know

  • The turbo model is not trained for translation; use medium or large for translating speech to English.
  • Model sizes range up to about 10 GB of VRAM for the large model, so larger models need capable hardware.
  • Accuracy and speed vary considerably by language.

Topics

#speech-recognition#speech-to-text#transcription#machine-learning#translation#python#pytorch

FAQ

What is whisper?
General-purpose speech recognition model that transcribes, translates and identifies the language of audio across many languages. Whisper is an open-source speech recognition model from OpenAI, trained on a large and diverse set of audio. It handles multilingual speech recognition, speech-to-English translation, and spoken language identification within a single Transformer sequence-to-sequence model, so it can replace several stages of a traditional speech processing pipeline.
Is whisper free to use? What license does it have?
whisper is open source under the MIT license. Check the license terms before using it in commercial products.
Is whisper still maintained?
Updated within 6 months. The last commit to whisper was on Aug 31, 2026.
What can I use whisper for?
Transcribing meeting recordings, interviews or podcasts into text. Adding subtitles or captions to audio and video content. Translating non-English speech into English text. Building voice input features into applications or local tools.
How do I install whisper?
Install ffmpeg using your system package manager (e.g. Homebrew on macOS): brew install ffmpeg Install Whisper from PyPI: pip install -U openai-whisper Transcribe an audio file from the command line: whisper audio.flac audio.mp3 audio.wav --model turbo
Where can I get whisper?
The source code and installation instructions are on GitHub: https://github.com/openai/whisper

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects