HuBERT
Self-supervised speech model learning from masked prediction of hidden units.
About
HuBERT (Hidden-Unit BERT) is a self-supervised speech representation model from Meta's FAIR lab, published in June 2021. It applies a BERT-style masked prediction loss over continuous speech, using an offline k-means clustering step to generate target labels without any transcripts.
By masking only the prediction regions, HuBERT learns a combined acoustic and language model from raw audio. It matched or beat wav2vec 2.0 on LibriSpeech and Libri-Light, with a 1B-parameter variant cutting word error rates by up to 19% on harder test sets.
The code and pre-trained models were released through fairseq, and HuBERT became a foundational technique for later speech models including WavLM.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 86; Trending score 62/100 from researcher assessment.