Research On Acoustic Source Localization And Speech Enhancement In Human-computer Interaction

Speech is the most natural way of human-computer interaction (HCI). There is neitherneed to touch or wear data devices, nor blinzone, while using it in the HCI. However, the speechsignal is easily contaminated by noise and other speakers around in the transmission, whichdeteriorating seriously the performance and efficiency of interaction.This paper mainly studiesthe method to improve the signal to noise ratio (SNR) of speech signal in the HCI system.The target acoustic source localization is the important part of multi-channel speechenhancement method based on the microphone array. In this paper, the target acoustic source islocated using the information of the time-difference of arrival (TDOA). In order to get moreaccurate TDOA, a new method for estimating TDOA is proposed, which filtering the noise byan appropriate threshold firstly and then estimating TDOA by the correlation of the signals.Simulation results show that this method is superior to the method of generalized crosscorrelation (GCC).To get more detailed information about the location of the sound source, based on themicrophone uniform linear array, a six-microphone’s parallel uniform linear array is advanced,which could position the target acoustic source in three-dimensional, combing with the methodof estimating TDOA proposed in this paper.On the basis of positioning the target acoustic source, the target speech enhancement isimplemented by the method of delay-and-sum beamforming. An improved solution about theweights in the fixed beamformer is presented to achieve better SNR improvement.The methods proposed in this paper is simulated in detail by the MATLAB experiments.The results show that in the context of SNR greater than1.5dB,the positioning accuracy of thetarget acoustic source can reach more than98%and improvement of SNRcan reach5dB, whileusing less number of microphones and easily implemented in hardware.