wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

We show for the first time that learning powerful representations from speech audio alone followed by fine-tuning on transcribed speech can outperform the best semi-supervised methods while being conceptually simpler. wav2vec 2.0 masks the speech input in the latent space and solves a contrastive task defined over a quantization of the latent representations which are jointly learned. Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/other test sets. When lowering the amount of labeled data to one hour, wav2vec 2.0 outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER. This demonstrates the feasibility of speech recognition with limited amounts of labeled data.

PDF Abstract NeurIPS 2020 PDF NeurIPS 2020 Abstract

Results from the Paper


 Ranked #1 on Speech Recognition on TIMIT (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Speech Recognition Libri-Light test-clean wav2vec 2.0 Large-10h-LV-60k Word Error Rate (WER) 2.5 # 1
Speech Recognition Libri-Light test-other wav2vec 2.0 Large-10h-LV-60k Word Error Rate (WER) 5.0 # 1
Speech Recognition LibriSpeech test-clean wav2vec 2.0 with Libri-Light Word Error Rate (WER) 1.8 # 9
Speech Recognition LibriSpeech test-other wav2vec 2.0 with Libri-Light Word Error Rate (WER) 3.3 # 6
Speech Recognition LibriSpeech test-other wav2vec 2.0 Word Error Rate (WER) 4.1 # 14
Speech Recognition TIMIT wav2vec 2.0 Percentage error 8.3 # 1

Methods