Realtime stereo localization of a moving speaker using filtered spectrogram correlation map algorithm

This paper describes a new method to localize a moving speaker using a stereo microphone tracking system. In the proposed system, signal generated by a moving speaker is received simultaneously by two microphones located on two ends of a 50Cm rod. Then, in order to find location of target a correlation based estimation process is performed to estimate the time difference between arriving signals to each microphone which is called TDOA (Time Difference of Arrival). TDOA estimation is done by peak detection in correlation curve of two received signals. Calculations are done in both time and frequency domain and are discussed by details. We also exploit some properties of speech and different kind of filter to make an improvement in results.