Near Field Source Localization In Complex Environments

Nowadays, acoustic source localization and speech enhancement usingmicrophone array are the hot research topics in speech signal processing. In the videoconferencing system, speech recognition system and hearing-aid devices, the positionof speaker can be determined by the acoustic source localization technique. However,the performance of existing acoustic source localization methods will be degradedseriously due to the environmental noise and room reverberation. In order to solvethe above problem, a robust Time Delay Estimation (TDE) method is proposed in thispaper, and a source localization system is constituted with the linear interpolationmethod. When this system is placed in front of the microphone array speechenhancement method as a pre-processing module, the background noise andreverberation can be suppressed more effectively, and better speech quality can beobtained in comparison with the reference methods.Three main parts are involved in the framework of the proposed acoustic sourcelocalization and speech enhancement system using microphone array, including TDE,source localization and speech enhancement. First, the time delay between the signalsfrom two microphones is estimated by the combined approach of statistical modelbased method and Acoustical Transfer Function (ATF) method. Then, the TDE is usedin the source localization module using linear interpolation method to determine thesource location. Finally, the source location is used as an input to the speechenhancement module to obtain the enhanced speech.In the proposed TDE algorithm, several improvements are made to theconventional ATF method. First, the Minima-Controlled Recursive Averaging (MCRA)method is adopted to estimate the noise intensity, and the Weighted EuclideanDistortion Measure (WEDM) based algorithm is used to remove the noise in thereceived signal from microphone array. Therefore, the effect of noise on the ATF canbe reduced. Then, the cross power spectrum of the two received signals is smoothedand whitened to remove the effect of reverberation on the ATF more effectively.Additionally, in order to improve the accuracy of TDE, Voice Activity Detection(VAD) is utilized to detect the noise periods which are useless in the calculation ofATF. The test results indicate that, the proposed TDE method outperforms theconventional methods in the complex environments with strong noise and severe reverberation.Generalized Sidelobe Canceller (GSC) that is a conventional microphone arrayspeech enhancement method is used as speech enhancement module. The evaluationof the enhanced speech is performed by Segmental Signal to Noise Ratio (SegSNR),Log-Spectral Distortion (LSD) and Perceptual Evaluation of Speech Quality (PESQ).The results indicate that, comparing with the reference methods, the proposed sourcelocalization method can provide localization of source accurately with better objectivespeech quality, while the SNR improvement remains acceptable.