Analyzing the Quality and Stability of a Streaming End-to-End On-Device Speech Recognizer
This addresses instability issues in incremental speech recognition for on-device applications, but it appears incremental as it focuses on analyzing and mitigating existing problems rather than proposing a new paradigm.
The paper tackles the instability problem in streaming end-to-end on-device speech recognizers, where partial results can be revised, by analyzing quality and stability, introducing new metrics, and exploring solutions to mitigate instability.
The demand for fast and accurate incremental speech recognition increases as the applications of automatic speech recognition (ASR) proliferate. Incremental speech recognizers output chunks of partially recognized words while the user is still talking. Partial results can be revised before the ASR finalizes its hypothesis, causing instability issues. We analyze the quality and stability of on-device streaming end-to-end (E2E) ASR models. We first introduce a novel set of metrics that quantify the instability at word and segment levels. We study the impact of several model training techniques that improve E2E model qualities but degrade model stability. We categorize the causes of instability and explore various solutions to mitigate them in a streaming E2E ASR system. Index Terms: ASR, stability, end-to-end, text normalization,on-device, RNN-T