CoVoST2: leaderboard

Speech-to-text translation over 2,880 hours of Mozilla Common Voice recordings from 78,000 speakers, covering 21 languages into English and English into 15 languages; Facebook AI (2020); BLEU.

Source: arxiv.org.

Interactive version: theaggregate.ai/benchmark?slug=covost2 · How It Works · Data refreshed daily, snapshot 2026-09-05.