Sound Source Localization And Speech Enhancement Techology Based On Microphone Array

Microphone array has been widely applied in various scenarios,such as video conferencing,homes,and automotive.As a means of acquiring and processing external signals,the microphone array is an indispensable part of smart devices.However,because of the environmental noise,reverberation,and other interference,the speech signal received by the smart device is not clear enough,diminishing the positioning accuracy,and the voice signal cannot be extracted.Therefore,for smart audio devices,locating the direction of the sound source and improve the sound quality have long been an important objective.Locating the direction of the sound source and using speech enhancement technology to separate the target speech signal from other interference signals is an indispensable part of the application of intelligent speech interaction.The main contents of this thesis are the sound source localization and speech enhancement technology based on microphone array.In terms of sound source localization,we design microphone array structure and studies the time delay estimation algorithm.In terms of speech enhancement,we improve the traditional dualchannel speech enhancement algorithm based on energy difference.At last,this thesis builds a hardware platform to implement the proposed algorithms.This thesis is organized as follows.First,we design the sound source positioning system of the six-element ring microphone array according to the actual environment and derive the relationship between time delay and sound source direction.Through simulation,we study the influence of different weighting functions on the generalized cross-correlation algorithm,and improves the generalized cross-correlation algorithm under weak reverberation environment.Then,we introduce and discuss the speech enhancement technology.The singlechannel spectrum subtraction and noise power spectrum estimation are introduced.The dual-channel speech enhancement algorithm based on energy difference is studied.The algorithm is optimized from the perspective of noise power spectrum estimation,and combined with post-masking processing to reduce the distortion of the speech signal.Finally,we build a hardware platform.We design and implement algorithms and systems,and test the feasibility of the platform.