Skip to main content
πŸ“– The AI Tool Bible
Whisper preview image
Whisper logo

Whisper

βœ“ Text source-checked Nov 11, 2026

Whisper is an automatic speech recognition system trained on 680,000 hours of multilingual and multitask supervised data.

Source-checked against the vendor's site Β· rebuilt from the vendor's own pagesProfile updated Link checked How we rate

FreeAudioWhisper large-v3 (not source-checked)8.6 / 10

In short

Whisper is an open-source speech recognition system trained on 680,000 hours of data. It offers robust multilingual transcription and translation using a Transformer architecture.

Whisper is an automatic speech recognition system trained on 680,000 hours of multilingual and multitask supervised data collected from the web. OpenAI open-sources the models and inference code to serve as a foundation for building useful applications and for further research on robust speech processing.

The Whisper architecture is a simple end-to-end approach implemented as an encoder-decoder Transformer. Input audio is split into 30-second chunks, converted into a log-Mel spectrogram, and then passed into an encoder. A decoder predicts the corresponding text caption, using special tokens to direct the model to perform language identification, phrase-level timestamps, multilingual speech transcription, and to-English speech translation.

The use of a large and diverse dataset leads to improved robustness to accents, background noise, and technical language. The system enables transcription in multiple languages and translation from those languages into English.

Whisper's code and model weights are released under the MIT License. There are six model sizes, four of which have English-only versions, offering speed and accuracy tradeoffs.

From the vendor's pages

According to Whisper: Whisper is an automatic speech recognition system trained on 680,000 hours of multilingual and multitask supervised data collected from the web. Its pages list this strength: Code and model weights are released under the MIT License. They also state this limitation: Whisper does not beat models that specialize in LibriSpeech performance.

β€” The AI Tool Bible editorial team

Pros

  • βœ… Code and model weights are released under the MIT License.
  • βœ… Models and inference code are open-sourced for building applications and research.
  • βœ… The model shows improved robustness to accents, background noise, and technical language.
  • βœ… The model supports transcription in multiple languages and translation into English.

Cons

  • ⚠️ Whisper does not beat models that specialize in LibriSpeech performance.

Facts

Platforms
web (not source-checked)
Open source
Yes (license: MIT)
GitHub stars
112,610 (checked Nov 13, 2026) (not source-checked)
Last GitHub push
Aug 31, 2026 (not source-checked)
First commit
Sep 2022 (not source-checked)
Company
OpenAI, Inc. (not source-checked)
Parent company
OpenAI (not source-checked)
Launched
Jan 1, 2015 (not source-checked)
Model used
Whisper large-v3 (not source-checked)
Package downloads (30 d)
4,986,176 Β· pypi openai-whisper Β· checked Oct 23, 2026 (not source-checked)

Where this tool sits

Use cases

Multilingual speech transcriptionTo-English speech translationLanguage identificationPhrase-level timestamps

Frequently asked

How much does Whisper cost to use?
Whisper is free. The models and inference code are open-sourced by OpenAI under the MIT License, allowing users to build applications without licensing fees.
What languages does Whisper support?
Whisper supports multilingual speech transcription and translation into English. It is trained on multilingual data, enabling it to handle various languages and accents with improved robustness.
Is Whisper accurate for noisy audio?
Yes. Training on a large and diverse dataset of 680,000 hours improves robustness to accents, background noise, and technical language, making it suitable for challenging audio environments.
Can Whisper provide timestamps for transcriptions?
Yes. The system uses special tokens to direct the model to perform phrase-level timestamps, allowing users to identify specific segments within the audio file.
What model sizes are available?
There are six model sizes available, with four having English-only versions. These options offer different speed and accuracy tradeoffs, allowing users to choose based on their specific performance needs.
See alternatives to Whisper β†’

Explore related

Reviews

The AI Tool Bible editorial teamEditor10/20/2026
Whisper: cheap baseline; compare OpenAI's newer API models and your VRAM

Pick Whisper over a managed transcription service when you have engineering time and want the model on your own hardware. The README says the code and model weights are released under the MIT License and lists six model sizes (tiny, base, small, medium, large and turbo) with approximate VRAM needs from about 1 GB up to about 10 GB for large. One concrete limitation from the README: the turbo model is not trained for translation tasks and returns the original language even if you request translate, so use medium or large when you need English output from other languages. The profile adds that there is no built-in diarisation and that silent segments can produce hallucinated text, so budget for a speaker-labelling step and some output cleanup. Before choosing the hosted route, compare options. OpenAI's whisper-1 model page lists $0.006 per minute, while OpenAI's pricing page also lists newer transcription models, including gpt-4o-mini-transcribe at $0.003 per minute and gpt-transcribe at $0.0045 per minute. If you need turnkey diarisation or streaming, the profile points to AssemblyAI instead.