Entity Retrieval (ER) in spoken dialog systems is a task that retrieves entities in a catalog for the entity mentions in user utterances. ER systems are susceptible to upstream errors, with Automatic Speech Recognition (ASR) errors being particularly troublesome. In this work, we propose a robust deep learning based ER system by leveraging ASR N-best hypotheses. Specifically, we evaluate different neural architectures to infuse ASR N-best through an attention mechanism. On 750 hours of audio data taken from live traffic, our best model achieves 11.07% relative error reduction while maintaining the same performance on rejecting out-of-domain ER requests.
Cite as: Wang, H., Chen, J., Laali, M., Durda, K., King, J., Campbell, W., Liu, Y. (2021) Leveraging ASR N-Best in Deep Entity Retrieval. Proc. Interspeech 2021, 261-265, doi: 10.21437/Interspeech.2021-1370
@inproceedings{wang21b_interspeech, author={Haoyu Wang and John Chen and Majid Laali and Kevin Durda and Jeff King and William Campbell and Yang Liu}, title={{Leveraging ASR N-Best in Deep Entity Retrieval}}, year=2021, booktitle={Proc. Interspeech 2021}, pages={261--265}, doi={10.21437/Interspeech.2021-1370} }