مکان یابی چند-منبعی در محیط های پرانعکاس توسط موسیقی-ریشه و خوشه بندی MULTI-SOURCE LOCALIZATION IN REVERBERANT ENVIRONMENTS BY ROOT-MUSIC AND CLUSTERING

ABSTRACT Localization of acoustic sources in reverberant environments by microphone arrays remains a challenging task in audio signal processing. As a matter of fact, most assumptions of commonly adopted models are not met in real applications. Moreover, in practical systems it is not convenient or possible to employ sophisticated and costly architectures, that require precise synchronization and fast data shuffling among sensors. In this paper, a new robust multi-step procedure for speaker localization in reverberant rooms is introduced and described. The new approach is based on a disturbed harmonics model of time delays in the frequency domain and employs the wellknown ROOT-MUSIC algorithm, after a preliminary distributed processing of the received signals. Candidate source positions are then estimated by clustering of raw TDOA estimates. Main features of the proposed approach, compared to previous solutions, are the capability of tracking multiple speakers and the high accuracy of the closed form TDOA estimator. ۱٫ INTRODUCTION Localization of acoustic sources in reverberant environments is an important task in many automatic systems for surveillance, videoconferencing, hands-free talking [I]. Spatial parameters obtained in the localization process can be used in a variety of applications: dereverberation of speech, fault prediction and analysis in machinery, cueing and tracking of TV cameras, speaker verification, etc. From a signal processing standpoint. the issue is a proper treatment of multiple arrivals, corresponding to both useful signal(s) and reflections. Reflective surfaces in closed environments are usually modeled by the introduction of virfual sources [2], whose number typically exceeds the microphone array size. This fact, coupled with the very large bandwidth of the signals of interest, makes unsuitable the parametric techniques used in narrow-band or moderately wide-band array processing in the presence of far-field sources [3][4][5]. For these reasons, most approaches to source localization involve the use of differential time delays (Time Delay of Arrival, TDOA) among pairs (“doublets”) of co-located microphones [6][7][8][9]. This process requires a joint parameter optimization from signals collected by many sensors at a time. Typically, TDOA estimation is performed by generalized crosscorrelation methods [6][9], that are appealing for their simplicity and ease of implementation. Anyway, generalized crosscorrelation methods assume a single-source model, which can be far from reality in many typical operating environments. A different model and strategies are thus needed to overcome the limitations of traditional approaches. From the point of view of system design, it is very important to reduce synchronization requirements and signal paths to a minimum, to reduce costs in current applications. In this work, we propose a novel three-stage strategy for the robust localization of multiple speakers in reverberant rooms. The first stage consists of data pre-whitening by use of Linear Predictive Coding (LPC). The effects of signal pre-whitening are to generate an approximate concentration of the likelihood function (under a simplifying Gaussian assumption) [7] and to reduce the reverberation effects (e.g. the number of significant TDOA to be estimated). In the second stage the TDOAs for the direct path and early (strongest) reflections are estimated by a closed-form parametric approach, based on the ROOT-MUSIC algorithm [ 121. Finally, the third stage finds the most likely position of the speakers by means of a clustering in space performed among all the estimated locations. The most dense clusters are selected as candidate speakers, thus eliminating most of false detections generated by outliers (virtual sources, localization ambiguities, impulsive noise, etc.).