HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units
Paper
•
2106.07447
•
Published
•
2
HuBERT
A collection of checkpoints from the HuBERT release, a speech encoder that learns powerful representations from unlabelled audio data.
Note The HuBERT paper, accepted at IEEE/ACM Transactions on Audio, Speech and Language Processing Volume 29.
facebook/hubert-base-ls960
Feature Extraction • Updated • 149k • 50Note The "base" HuBERT model fine-tuned on 960 hours of LibriSpeech ASR data.
facebook/hubert-large-ll60k
Feature Extraction • Updated • 7k • 25Note The "large" HuBERT model pre-trained on LibriVox 60k hours.
facebook/hubert-large-ls960-ft
Automatic Speech Recognition • Updated • 1.45M • 63Note A fine-tuned version of hubert-large-ll60k, fine-tuned on 960 hours of LibriSpeech ASR data.
facebook/hubert-xlarge-ll60k
Feature Extraction • Updated • 1.14k • 5Note The "extra-large" HuBERT model pre-trained on LibriVox 60k hours.
facebook/hubert-xlarge-ls960-ft
Automatic Speech Recognition • Updated • 5.94k • 12Note A fine-tuned version of hubert-xlarge-ll60k, fine-tuned on 960 hours of LibriSpeech ASR data. This is the most performant HuBERT checkpoint in the release, achieving a WER of 1.8/2.9% on the LibriSpeech test clean/other subsets respectively.