Modular Commands Recognition Based on Semantic Understanding

Recent commands recognition methods based on deep neural networks (DNNs) have achieved remarkable results, but there are still generalization problems in real scenes. Most commands recognition models can only recognize the command in a fixed format, which is not convenient. In this research, we propose a modular model that separates speech recognition model which can make full use of advanced end-to-end ASR and semantic recognition model which is designed for semantic understanding. We use the tokens recognized by speech recognition model as the input of the semantic recognition model. We tested the performance of the model by using both fixed form commands and commands in the daily pragmatic environment. For this reason, we specially build a command dataset in the daily pragmatic environment. Experimental results show that the model proposed in this paper not only performs well in the task of fixed-format commands recognition but can also realize efficient and accurate command recognition in daily pragmatic environment.