Two Stage Contextual Word Filtering for Context bias in Unified Streaming and Non-streaming Transducer
This work addresses the challenge of context bias in ASR for better recognition of rare words, representing an incremental improvement in domain-specific efficiency.
The paper tackles the problem of recognizing infrequent words like entities in end-to-end automatic speech recognition by proposing a two-stage contextual word filtering method that improves accuracy and speeds up inference. Experiments show over 20% CER reduction compared to baseline and stabilize RTF within 0.15 for large contextual lists.
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.