A SOUND-SOURCE TRACKING DEVICE TO TRACK MULTIPLE TALKERS FROM MICROPHONE ARRAY AND LAVALIER MICROPHONE DATA

Many algorithms have been developed to localize audio signals based on the differences in the sound as it arrives at spatially disparate microphones in a larger array or arrays of microphones. While many of these systems perform well with one sound source, tracking multiple sound sources in parallel remains to be a real challenge. In the project, which is presented here, the task was to localize talkers, and then reproduce their voices – which were recorded at close distance with lavalier microphones – spatially correct using a loudspeaker rendering system. The localization process was based on time delay differences between various channels of a small-aperture pyramidal five-microphone array. In addition to this common practice, the information gained from the presence of the talker-worn microphones was utilized to estimate the signal-to-noise ratio (SNR) between each talker and the concurrent talkers. An algorithm was designed to select time-frequency bins that showed a high SNR for robust localization of the various talkers and to identify the talkers of the localized sources. It was found that correlating the talker-worn microphones with the microphone array allows for a greater accuracy and precision of localization than with only the microphone array