Post-decoder Biasing for End-to-End Speech Recognition of Multi-turn Medical Interview
This addresses the challenge of handling decoding shifts and rare words in ASR for medical applications, though it is incremental as it builds on existing E2E methods.
The paper tackles the problem of recognizing domain-specific rare words in end-to-end speech recognition, particularly for multi-turn medical interviews, by proposing a post-decoder biasing method that achieves relative improvements of 9.3% and 5.1% for rare word subsets.
End-to-end (E2E) approach is gradually replacing hybrid models for automatic speech recognition (ASR) tasks. However, the optimization of E2E models lacks an intuitive method for handling decoding shifts, especially in scenarios with a large number of domain-specific rare words that hold specific important meanings. Furthermore, the absence of knowledge-intensive speech datasets in academia has been a significant limiting factor, and the commonly used speech corpora exhibit significant disparities with realistic conversation. To address these challenges, we present Medical Interview (MED-IT), a multi-turn consultation speech dataset that contains a substantial number of knowledge-intensive named entities. We also explore methods to enhance the recognition performance of rare words for E2E models. We propose a novel approach, post-decoder biasing, which constructs a transform probability matrix based on the distribution of training transcriptions. This guides the model to prioritize recognizing words in the biasing list. In our experiments, for subsets of rare words appearing in the training speech between 10 and 20 times, and between 1 and 5 times, the proposed method achieves a relative improvement of 9.3% and 5.1%, respectively.