Dissecting User-Perceived Latency of On-Device E2E Speech Recognition

TL;DR

This paper investigates how different factors like model architecture and endpointing affect user-perceived latency in on-device end-to-end speech recognition systems, highlighting that traditional measures may not accurately reflect real-world latency.

Contribution

It provides an analysis of various techniques impacting latency, emphasizing the importance of token emission and endpointing over conventional computational metrics.

Findings

Model size and FLOPS are not strongly correlated with user-perceived latency.

Token emission timing and endpointing significantly influence latency.

Joint ASR and endpointing with alignment regularization offers optimal latency-WER trade-offs.

Abstract

As speech-enabled devices such as smartphones and smart speakers become increasingly ubiquitous, there is growing interest in building automatic speech recognition (ASR) systems that can run directly on-device; end-to-end (E2E) speech recognition models such as recurrent neural network transducers and their variants have recently emerged as prime candidates for this task. Apart from being accurate and compact, such systems need to decode speech with low user-perceived latency (UPL), producing words as soon as they are spoken. This work examines the impact of various techniques - model architectures, training criteria, decoding hyperparameters, and endpointer parameters - on UPL. Our analyses suggest that measures of model size (parameters, input chunk sizes), or measures of computation (e.g., FLOPS, RTF) that reflect the model's ability to process input frames are not always strongly correlated with observed UPL. Thus, conventional algorithmic latency measurements might be inadequate in accurately capturing latency observed when models are deployed on embedded devices. Instead, we find that factors affecting token emission latency, and endpointing behavior have a larger impact on UPL. We achieve the best trade-off between latency and word error rate when performing ASR jointly with endpointing, while utilizing the recently proposed alignment regularization mechanism.