Speaker Tracking and Identifying Based on Indoor Localization System and Microphone Array

This paper presents a novel multimodal system to track the participants and identify the active speaker in the smart meeting room. Indoor localization system, Cicada, is used to obtain the location and identity information of the participants. Cicada, developed by our research group, can provide much higher localization precision than most visual tracking systems available. Four eight-element microphone arrays are used to estimate TDOA (Time Difference of Arrival) of the audio signals, which are indicators of the location of the speaker. Integrating these two cues from Cicada and microphone arrays into a statistical framework, results in a more robust solution to speaker localization and identification. The experiments show that the average error in the participant location estimation is about 15 cm and the system can achieve approximately 90% accuracy in speaker identification on average.