Application of Frequency Domain Binaural Model for Keyword Spotting on Simultaneous Speech

摘要

In this paper, we have presented the method ofemploying Frequency Domain Binaural Model(FDBM) for Keyword Spotting System (KWS).By employing FDBM, we could address theproblem of using KWS in the concurrent speechsituations and increase the word detection rate ofthe system for an interfered signal which isshown by the word detection rate improvement of49.05% for the speech from 0° azimuth and31.8% for other directions.We also confirmed that by using FDBM, wecould reduce the false positive rate of the KWSmodel by attenuating the interfering speech.