Improved Spoken Language Representation for Intent Understanding in a Task-Oriented Dialogue System

Abstract

Successful applications of deep learning technologies in the natural language processing domain have improved text-based intent classifications. However, in practical spoken dialogue applications, the users’ articulation styles and background noises cause automatic speech recognition (ASR) errors, and these may lead language models to misclassify users’ intents. To overcome the limited performance of the intent classification task in the spoken dialogue system, we propose a novel approach that jointly uses both recognized text obtained by the ASR model and a given labeled text. In the evaluation phase, only the fine-tuned recognized language model (RLM) is used. The experimental results show that the proposed scheme is effective at classifying intents in the spoken dialogue system containing ASR errors.

Unsupervised Representation Learning with Task-Agnostic Feature Masking for Robust End-to-End Speech Recognition

“…To make matters worse, since most of the benchmark curated speech datasets [2][3][4][5] are built with very limited diversity, mostly representing healthy adults, it is challenging to accurately recognize the speech of children [9,10], the elderly [11][12][13], or those using dialects. Consequently, speech recognition performance suffers in highly variable scenarios, such as far-field or noisy environments [6][7][8]14,15], where the conditions or personal characteristics [16][17][18][19] degrade the performance compared with normal speech. In addition, recognizing new or trending words is important for ASR systems, but updating already built end-to-end ASR systems every time is time-consuming and resource-intensive.…”