Back to whisper
OpenAI /

Whisper large-v3
Whisper large-v3

Open WeightsAudio & SpeechaudioUpdated November 6, 2023

Whisper Large v3: OpenAI's Multilingual Speech Recognition Model

Model Overview

Whisper Large v3 is OpenAI's latest and most capable automatic speech recognition (ASR) model, released in November 2023. Trained on 5 million hours of labeled audio data spanning 100+ languages, Whisper Large v3 achieves state-of-the-art word error rates across a broad range of languages and accents. It supports multilingual speech recognition, speech translation (to English), language identification, and voice activity detection.


✨ Key Features

Specification Table
FeatureDescription
Multilingual ASRTranscription in 100+ languages with state-of-the-art accuracy
Speech TranslationDirect translation of speech from any supported language to English
5M Hours TrainingTrained on the largest labeled audio dataset used for ASR
Long-Form AudioHandles audio files of arbitrary length via sliding window chunking
Open WeightsReleased as open-source on HuggingFace (Apache 2.0)

📊 Benchmarks (Word Error Rate — lower is better)

Specification Table
LanguageWER
English~2.7%
Spanish~3.0%
French~4.2%
German~3.8%
Chinese~7.5%

🔗 Resources


📜 License

MIT License — Fully open-source, free for commercial and non-commercial use.

Key Features

Supports speech-to-text transcription and translation

Feature 01

Trained on 1M hours of weakly labeled audio data

Feature 02

Highly accurate across 99+ languages

Feature 03

Permissive MIT open-source license

Feature 04

You might also want to compare

Verified Sources

Tags

audiospeech-to-textopen-source

Model Specs

open-weights

Parameters

1.55B

Context Window

30s

License

MIT

Deployment

self-hostableapi-only

Resources & Links

Lineage

Model Family

Part of the whisper family

Curator Notes

Released during OpenAI DevDay. Uses 128 mel frequency bins and processes audio in 30-second sliding windows.

Compare Specs

Compare parameters, context windows, modalities, and benchmark scores of this model side-by-side with others.

Compare Model