On Scaling Contrastive Representations for Low-Resource Speech Recognition

Abstract

Recent advances in self-supervised learning through contrastive training have shown that it is possible to learn a competitive speech recognition system with as little as 10 minutes of labeled data. However, these systems are computationally expensive since they require pre-training followed by fine-tuning in a large parameter space. We explore the performance of such systems without fine-tuning by training a stateof-the-art speech recognizer on the fixed representations from the computationally demanding wav2vec 2.0 framework. We find performance to decrease without fine-tuning and, in the extreme low-resource setting, wav2vec 2.0 is inferior to its predecessor. In addition, we find that wav2vec 2.0 representations live in a low dimensional subspace and that decorrelating the features of the representations can stabilize training of the automatic speech recognizer. Finally, we propose a bidirectional extension to the original wav2vec framework that consistently improves performance.

Improving Hybrid CTC/Attention End-to-end Speech Recognition with Pretrained Acoustic and Language Model

“…During fine-tuning, we only update the parameters of the Transformer based context network following the wav2vec2.0 [5]. Fine-tuning wav2vec2.0 on labeled data with CTC objective [16] has been well verified [30,31,32]. However, according to the work [6], the results of fine-tuning wav2vec2.0 based on a vanilla Transformer S2S ASR model with cross-entropy criterion can only achieve a very limited result.…”