HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units

HuBERT: самообучение представлений речи посредством маскированного предсказания скрытых единиц
Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed
2021-01-01

HuBERThidden unitsk-means clusteringmasked predictionself-supervised speech representation learning
Self-supervised approaches for speech representation learning are challenged by three unique problems: (1) there are multiple sound units in each input utterance, (2) there is no lexicon of input sound units during the pre-training phase, and (3) sound units have variable lengths with no explicit segmentation. To deal with these three problems, we propose the Hidden-Unit BERT (HuBERT) approach for self-supervised speech representation learning, which utilizes an offline clustering step to provide aligned target labels for a BERT-like prediction loss. A key ingredient of our approach is applying the prediction loss over the masked regions only, which forces the model to learn a combined acoustic and language model over the continuous inputs. HuBERT relies primarily on the consistency of the unsupervised clustering step rather than the intrinsic quality of the assigned cluster labels. Starting with a simple k-means teacher of 100 clusters, and using two iterations of clustering, the HuBERT model either matches or improves upon the state-of-the-art wav2vec 2.0 performance on the Librispeech (960h) and Libri-light (60,000h) benchmarks with 10min, 1h, 10h, 100h, and 960h fine-tuning subsets. Using a 1B parameter model, HuBERT shows up to 19% and 13% relative WER reduction on the more challenging dev-other and test-other evaluation subsets.
1
A 1B-parameter HuBERT model achieves up to 19% relative WER reduction on dev-other and 13% on test-other.
2
Applying the prediction loss only to masked regions encourages learning a joint acoustic and language model over continuous speech inputs.
3
HuBERT addresses multiple sound units, absent pretraining lexicons, and variable unit lengths using offline clustering to generate aligned targets for masked prediction.
4
HuBERT depends primarily on consistency of unsupervised cluster assignments rather than the intrinsic quality of cluster labels.
5
With a 100-cluster k-means teacher and two clustering iterations, HuBERT matches or exceeds wav2vec 2.0 on Librispeech and Libri-light across multiple fine-tuning data sizes.

speech representations learned from unlabeled continuous speech inputs

the masked-prediction learning behavior and recognition performance of the representations, including acoustic–language modeling and WER

Publication Details
Publication Date
2021-01-01
Journal
IEEE/ACM Transactions on Audio Speech and Language Processing
Publisher
Institute of Electrical and Electronics Engineers
ISSN
2329-9290
Cited by
5978
Access Type
Author Information
Authors
Wei-Ning Hsu
Benjamin Bolte
Yao-Hung Hubert Tsai
Kushal Lakhotia
Ruslan Salakhutdinov
Abdelrahman Mohamed
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%