Fact · Speech recognition
4.8percent word error rate
With just ten minutes of labeled data and pre-training on 53,000 hours of unlabeled audio, wav2vec 2.0 still reached 4.8 percent word error rate on LibriSpeech clean and 8.2 percent on other.
Checked against its source on October 16, 2026
In the source's own words
“Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER.”
wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations, Alexei Baevski et al., arXiv (NeurIPS 2020), 2020, Abstract.
Open the original- What it measures
- Self-supervised pre-training lets a model learn the sounds of a language from audio alone, so little transcribed speech is needed afterwards.
- What it does not show
- English LibriSpeech with a language model; the tiny labeled set is a research setup, not how products are built.