Studies on binaural and monaural signal analysis methods and applications

摘要

Sound signals can contain a lot of information about the environment and the sound sources present in it. This thesis presents novel contributions to the analysis of binaural and monaural sound signals. Some new applications are introduced in this work, but the emphasis is on analysis methods. The three main topics of the thesis are computational estimation of sound source distance, analysis of binaural room impulse responses, and applications intended for augmented reality audio. A novel method for binaural sound source distance estimation is proposed. The method is based on learning the coherence between the sounds entering the left and right ears. Comparisons to an earlier approach are also made. It is shown that these kinds of learning methods can correctly recognize the distance of a speech sound source in most cases. Methods for analyzing binaural room impulse responses are investigated. These methods are able to locate the early reflections in time and also to estimate their directions of arrival. This challenging problem could not be tackled completely, but this part of the work is an important step towards accurate estimation of the individual early reflections from a binaural room impulse response. As the third part of the thesis, applications of sound signal analysis are studied. The most notable contributions are a novel eyes-free user interface controlled by finger snaps, and an investigation on the importance of features in audio surveillance. The results of this thesis are steps towards building machines that can obtain information on the surrounding environment based on sound. In particular, the research into sound source distance estimation functions as important basic research in this area. The applications presented could be valuable in future telecommunications scenarios, such as augmented reality audio.