Advancing Connectionist Temporal Classification With Attention Modeling

TL;DR

This paper enhances all-neural speech recognition by integrating advanced attention mechanisms into the CTC framework, leading to significant improvements in word error rates on a large-scale voice recognition task.

Contribution

It introduces novel attention modeling techniques within CTC, including context vectors with time convolution features and vector-based attention weights, improving speech recognition accuracy.

Findings

Achieved about 20% relative reduction in word error rates.

Demonstrated effectiveness on a 3400-hour Microsoft Cortana dataset.

Outperformed baseline models consistently.

Abstract

In this study, we propose advancing all-neural speech recognition by directly incorporating attention modeling within the Connectionist Temporal Classification (CTC) framework. In particular, we derive new context vectors using time convolution features to model attention as part of the CTC network. To further improve attention modeling, we utilize content information extracted from a network representing an implicit language model. Finally, we introduce vector based attention weights that are applied on context vectors across both time and their individual components. We evaluate our system on a 3400 hours Microsoft Cortana voice assistant task and demonstrate that our proposed model consistently outperforms the baseline model achieving about 20% relative reduction in word error rates.