Online Automatic Speech Recognition with Listen, Attend and Spell Model

TL;DR

This paper analyzes the limitations of attention-based LAS models in online speech recognition and proposes a simple technique to achieve fully online recognition with high accuracy and low latency, validated in a large-scale deployment.

Contribution

It introduces a novel method enabling fully online LAS speech recognition with improved latency and accuracy, validated through a production deployment.

Findings

Achieves within 4% relative CER of offline LAS in online mode

Operates at 12% lower latency than hybrid models

First large-scale deployment of a fully online LAS model

Abstract

The Listen, Attend and Spell (LAS) model and other attention-based automatic speech recognition (ASR) models have known limitations when operated in a fully online mode. In this paper, we analyze the online operation of LAS models to demonstrate that these limitations stem from the handling of silence regions and the reliability of online attention mechanism at the edge of input buffers. We propose a novel and simple technique that can achieve fully online recognition while meeting accuracy and latency targets. For the Mandarin dictation task, our proposed approach can achieve a character error rate in online operation that is within 4% relative to an offline LAS model. The proposed online LAS model operates at 12% lower latency relative to a conventional neural network hidden Markov model hybrid of comparable accuracy. We have validated the proposed method through a production scale deployment, which, to the best of our knowledge, is the first such deployment of a fully online LAS model.