Whisper

Whisper is an open-source automatic speech recognition model from OpenAI that transcribes spoken audio into text, translates non-English speech into English, and identifies the language being spoken.

Released in September 2022 under the MIT license, Whisper is a Transformer encoder-decoder model originally trained on 680,000 hours of labeled audio. It ships in sizes from tiny at 39 million parameters to large at 1.55 billion, plus a faster turbo variant that drops translation support.

The ladder below steps through those checkpoints with the speed and memory figures OpenAI publishes for each one.

Interactive diagram — enable JavaScript to view.

Python developers install it from PyPI with python -m pip install -U openai-whisper and the ffmpeg command-line tool. Transcription takes two calls, whisper.load_model() then .transcribe(), and both run locally on CPU or GPU. OpenAI also hosts it in its audio API as whisper-1, now legacy beside the newer gpt-transcribe and gpt-live-transcribe. OpenAI still points to it for word-level timestamps, subtitle formats, and English translation.

In coding workflows, Whisper mostly shows up inside dictation front ends, such as superwhisper, that can run it on device and feed the resulting text straight into editors like Cursor. The whisper.cpp and faster-whisper reimplementations trade the reference Python code for lower-latency, lower-memory inference. Trained on weakly labeled data, the model can also output text that was never spoken.

Official website: github.com/openai/whisper

Python Speech Recognition

Tutorial

The Ultimate Guide To Speech Recognition With Python

An in-depth tutorial on speech recognition with Python. Learn which speech recognition library gives the best results and build a full-featured "Guess The Word" game with it.

advanced data-science machine-learning

For additional information on related topics, take a look at the following resources:

By Martin Breuss • Updated Aug. 23, 2026
AI Coding Tools Share Feedback