Unimodal Aggregation for CTC-based Speech Recognition

TL;DR

This paper introduces a unimodal aggregation method for CTC-based speech recognition that improves feature representation, reduces sequence length, and enhances accuracy and efficiency in non-autoregressive models.

Contribution

The paper proposes a novel unimodal aggregation technique that segments and integrates feature frames for better text token representation in CTC-based speech recognition.

Findings

UMA improves recognition accuracy on Mandarin datasets.

UMA reduces computational complexity compared to regular CTC.

Integrating self-conditioned CTC further enhances performance.

Abstract

This paper works on non-autoregressive automatic speech recognition. A unimodal aggregation (UMA) is proposed to segment and integrate the feature frames that belong to the same text token, and thus to learn better feature representations for text tokens. The frame-wise features and weights are both derived from an encoder. Then, the feature frames with unimodal weights are integrated and further processed by a decoder. Connectionist temporal classification (CTC) loss is applied for training. Compared to the regular CTC, the proposed method learns better feature representations and shortens the sequence length, resulting in lower recognition error and computational complexity. Experiments on three Mandarin datasets show that UMA demonstrates superior or comparable performance to other advanced non-autoregressive methods, such as self-conditioned CTC. Moreover, by integrating self-conditioned CTC into the proposed framework, the performance can be further noticeably improved.