Intention understanding based on multi-source information integration for Chinese Mandarin spoken commands

In this paper, we address the problem of Intention understanding for Chinese Mandarin spoken commands. Unlike the previous works, we propose an intention understanding approach including not only the detection of command content, but also the detection of user´s affective state. For command content detection, we propose a strategy of keyword combinations analysis using concept restrictions, based on N-best speech recognition results. For affective state detection, we propose a method of multi-source information integration in decision level, using a weighted maximum confidence score to get a high reliability for both acoustic features and speech recognized text. Experimental results show satisfactory effectiveness in command content detection, while combining multi-source information improves the performance of affective state detection.