Research On Passive Acoustic Localization Algorithm

Speech signal processing is a traditional as well as modernized research field. Many scholars have made great progress and achievements in the area of speech signal processing, especially in the fields of speech recognition, speech enhancement, sound source localization which have been widely applied in real life, been used to develop a large number of practical products and made a great contribution to our daily life. Although sound source localization technology has become a heated research field in recent years and become a multi-direction and multi-method technology, it is still a practical technology. In other words, it is a process of speech signal. Therefore, this thesis takes sound source localization as a starting point and mainly focuses on studying a new method of indoor sound source location.In our daily life, voice signals are complex and usually unstable, nonlinear and random. Especially in certain local indoor environment, due to the interference of reflection and reverberation, there exist great limitations and errors in using traditio nal signal processing method. So, this thesis proposes a method of the speech signa 1 processing which is based on Hilbert Huang transform (HHT) and establishes a bi-dimensional indoor microphone array model to locate indoor sound source by com bining the methods of delay estimation and wavelet pre-processing method.HHT, as a method for time-frequency localization analysis, takes the signal itself as the starting point and generates an inherent basis function self-adaptively rather than artificially set. In addition, compared to other methods of signal processing, HHT is more accurate and detailed in processing signals. Based on such an advantage, while processing signals, this method can separate signals from interfering noise in actual environment and thus greatly reduces the influence of indoor reverberation on the accuracy of positioning system, which enables this method to have obviously great advantage in processing unstable, nonlinear speech signal.Then, this thesis puts forward an innovative method, which is the positioning algorithm of time delay estimation based on HHT. Such a method features in using an important method of empirical mode decomposition (EMD) in HHT to decompose the collected speech signal and then filter out each component that is defined as the inherent mode functions or intrinsic mode function (IMF). Such a component, being self-adaptive, can fully demonstrate the characteristics of speech signal. Then the wavelet soft threshold is used to denoise and reconstruct the obtained IMF components, obtain the time delay estimation by combining the correlation theory and get the sound source azimuth angle by using the collection interpolation method. The localization of target sound source can be achieved in experiment, which proves the feasibility of this method.