Two Stage Contextual Word Filtering for Context bias in Unified Streaming and Non-streaming Transducer

TL;DR

This paper introduces a two-stage filtering method for contextual words in a unified streaming and non-streaming end-to-end speech recognition system, significantly improving accuracy and efficiency.

Contribution

It presents a novel approach to generate high-quality contextual lists using phone-level streaming output, enhancing recognition accuracy and inference speed.

Findings

Over 20% CER reduction compared to baseline

RTF stabilized within 0.15 with 6,000+ contextual words

Effective for both streaming and non-streaming ASR

Abstract

It is difficult for an E2E ASR system to recognize words such as entities appearing infrequently in the training data. A widely used method to mitigate this issue is feeding contextual information into the acoustic model. Previous works have proven that a compact and accurate contextual list can boost the performance significantly. In this paper, we propose an efficient approach to obtain a high quality contextual list for a unified streaming/non-streaming based E2E model. Specifically, we make use of the phone-level streaming output to first filter the predefined contextual word list then fuse it into non-casual encoder and decoder to generate the final recognition results. Our approach improve the accuracy of the contextual ASR system and speed up the inference process. Experiments on two datasets demonstrates over 20% CER reduction comparing to the baseline system. Meanwhile, the RTF of our system can be stabilized within 0.15 when the size of the contextual word list grows over 6,000.