Speech Recognition2019multilingual

Mozilla Common Voice

Massive multilingual dataset of transcribed speech. Covers diverse demographics and accents. Over 100 languages, updated continuously by Mozilla Foundation.

Metrics:wer

Paper / Website Download

Current State of the Art

Whisper Large V3

OpenAI

8.4

wer

Common Voice — wer

3 results · 2 SOTA advances · lower is better

All results

SOTA frontier

wer Progress Over Time

Showing 2 breakthroughs from Jun 2020 to Feb 2025

Key Milestones

Jun 2020

wav2vec 2.0 Large (960h)

WER (%) on Common Voice 9 English. Source: Papers With Code / wav2vec2 model card

10.5

Source

Feb 2025

Whisper Large V3Current SOTA

WER (%) on Common Voice 15 English test set. Source: B-Whisper paper Table 1 baseline, arxiv:2502.11572

8.4

-20.0%

Source

Total Improvement

20.0%

Time Span

4y 9m

Breakthroughs

Current SOTA

8.4

Top Models Performance Comparison

Top 3 models ranked by wer (lower is better)

Best Score

8.4

Top Model

Whisper Large V3

Models Compared

Score Range

2.8

werPrimary

#	Model	Score	Paper / Code	Date
1	Whisper Large V3Open Source OpenAI	8.4	Robust Speech Recognition via Large-Scale Weak Supervision (Whisper)	Dec 2022
2	wav2vec 2.0 Large (960h)Open Source Meta AI	10.5	wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations	Jun 2020
3	Whisper Large-v2Open Source OpenAI	11.2	Robust Speech Recognition via Large-Scale Weak Supervision (Whisper)	Dec 2022

Dec 2022Models: Whisper Large-v2, Whisper Large V3

Paper

wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

Jun 2020Models: wav2vec 2.0 Large (960h)