Major Release OpenAI Audio MIT

OpenAI Whisper v3

State-of-the-art open speech recognition — 99 languages, open weights

Released 2023-11-06 · ~1.55B

Overview

OpenAI's multilingual speech recognition system

Specifications

DeveloperOpenAI
Release date2023-11-06
Model typeAudio
Parameters~1.55B
ArchitectureTransformer encoder-decoder
LicenseMIT
Input modalitiesaudio
Output modalitiestext

Benchmarks

Benchmark Whisper v3
WER English 5.1%
Language Coverage 99 languages
Real-time Factor 0.8x
WER (English) ~2.7%

Availability

Open Source OpenAI API Self-hosted (HuggingFace) Groq

Official references