Speech

Whisper

OpenAI's free speech-to-text model that transcribes and translates audio in dozens of languages, offline.

Rank

#6

Stars

112k

Licence

MIT

Checked

6 Nov 2026

What it does

Whisper turns speech into text. OpenAI trained it on 680,000 hours of audio from the web, so it copes with accents, background noise and technical words better than older open tools. It can also translate speech in other languages into English text. It runs on your own machine, and a GPU makes it much faster.

How to use it

Shell: PowerShell (Windows 11, Python 3.9+ installed)

1. 1

   Install ffmpeg, which Whisper uses to read audio files.

   winget install Gyan.FFmpeg

2. 2

   Install Whisper with pip (a virtual environment is best).

   pip install -U openai-whisper

3. 3

   Transcribe a file. It writes .txt, .srt and .vtt files next to it.

   whisper meeting.mp3 --model turbo --language English

4. 4

   Translate Hindi speech into English text.

   whisper interview.mp3 --model medium --language Hindi --task translate

Best practices

• Always have a person check transcripts that matter. Whisper can invent words that were never spoken, especially in silences.
• Set --language when you know it; auto-detect can guess wrong on short clips.
• Use the smaller models (tiny, base) on a CPU-only laptop and turbo or large on a GPU.
• Get consent before transcribing other people's voices, and keep sensitive audio offline.

In the news

• TechCrunch · 21 Sept 2022 · OpenAI open-sources Whisper, a multilingual speech recognition system
• AP News · 26 Oct 2024 · Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said