TL;DR
This paper introduces a single-stage training approach for ASR that combines unsupervised masked CPC and supervised CTC losses, enabling effective use of unlabeled and labeled data simultaneously.
Contribution
It proposes a novel joint training method that integrates contrastive and CTC losses in a single stage, simplifying the training pipeline for speech recognition models.
Findings
Achieves similar WER to wav2vec 2.0 on Librispeech 100-hour dataset
Demonstrates effective utilization of unlabeled data in supervised ASR training
Suggests contrastive learning acts as regularization for CTC loss
Abstract
Self-supervised learning (SSL) has shown promise in learning representations of audio that are useful for automatic speech recognition (ASR). But, training SSL models like wav2vec~2.0 requires a two-stage pipeline. In this paper we demonstrate a single-stage training of ASR models that can utilize both unlabeled and labeled data. During training, we alternately minimize two losses: an unsupervised masked Contrastive Predictive Coding (CPC) loss and the supervised audio-to-text alignment loss Connectionist Temporal Classification (CTC). We show that this joint training method directly optimizes performance for the downstream ASR task using unsupervised data while achieving similar word error rates to wav2vec~2.0 on the Librispeech 100-hour dataset. Finally, we postulate that solving the contrastive task is a regularization for the supervised CTC loss.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
MethodsInfoNCE · Contrastive Predictive Coding
