Direction of arrival estimation for indoor environments based on acoustic composition model with a single microphone. (September 2022)

• Record Type: Journal Article
• Authors: Guo, Xingchen; Xu, Xuexin; Chen, Xunquan; Chen, Jinhui; Jia, Rong; Zhang, Zhihong; Takiguchi, Tetsuya; Hancock, Edwin R.
• Is Part Of: Pattern recognition. Volume 129(2022)
• Publication Date: 2022-09
• Subjects: Gaussian mixture model (GMM) -- Acoustic transfer function (ATF) -- Talker localization; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108715
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: This paper presents an effective method for multiple talker localisation using only a single microphone in a room. In this study, we estimate the acoustic transfer function from reverberant speech, using clean speech model. We process the speech signal in the cepstrum domain, and propose Composite Reverberant Speech (CRS) model and Direct Training Reverberant Speech (DTRS) model to obtain reverberant speech model. Abstract: This paper presents an effective method for multiple talker localisation using only a single microphone in a room. One of the main challenge here is obtaining a model that can be used for estimating the localization parameter. This model must be sensitive to all possible speaker locations and correctly discriminate their positions. The reverberant speech signal in a room environment can be composited by the clean speech and the acoustic transfer function (ATF). The ATF is a useful tool to describe changes of the speech source, and the approaches based on ATF can thus be used to identify talker localizations with a single microphone. This paper presents two methods, referred to as Composite Reverberant Speech (CRS) model and Direct Training Reverberant Speech (DTRS) model, and uses these methods for obtaining the ATF of a room. The approaches based on proposed methods can successfully and accurately process multi-talker localization task with single microphone. Experiments also demonstrate the effectiveness of the proposed methods.