Abstract
In this paper, an improved distributed unscented Kalman particle filter (DUKPF) is proposed for the problem of tracking a single moving acoustic source in noisy and reverberant environments with distributed microphone networks. The conventional DUKPF employs the unscented Kalman filter (UKF) for its proposal of particle sampling, whereas the UKF incorporates one single observation from a certain localization function, which is vulnerable to noise or reverberation. To alleviate this problem, multiple observations are extracted from the localization function at each node and incorporated into the state update of the UKF via the probability data association (PDA) technique, yielding the PDA-UKF. Next, employing the PDA-UKF for the proposal of particle sampling, the improved DUKPF is further developed. Finally, the improved DUKPF is adapted for the acoustic source tracking problem, and a distributed acoustic source tracking method is presented. Simulation results reveal that the improved DUKPF achieved better tracking performance than the conventional DUKPF in different noisy and reverberant conditions.
Highlights
The problem of acoustic source localization and tracking has played an important role in many applications, such as security monitoring systems, human-machine interaction, speech enhancement, and far-field speech recognition systems increasingly pervasive in recent years [1]–[4]
The time difference of arrivals (TDOAs) observations are first incorporated into the probability data association (PDA)-unscented Kalman filter (UKF) for the proposal of particle filtering, and exploited by the multihypothesis model for the local likelihood of the improved distributed unscented Kalman particle filter (DUKPF)
To alleviate the adverse effects of noise and reverberation, multiple TDOA observations are extracted at each node and incorporated into the state update of the UKF through the PDA technique, yielding the probability data association unscented Kalman filter (PDA-UKF)
Summary
The problem of acoustic source localization and tracking has played an important role in many applications, such as security monitoring systems, human-machine interaction, speech enhancement, and far-field speech recognition systems increasingly pervasive in recent years [1]–[4]. Zhong and Hopgood [17] combined EKF with PF to estimate the time-varying speaker position, which exploits the amplitude of the TDOA observations for the innovation of EKF Overall, these approaches belong to the centralized methods, i.e., all nodes’ data are collected and transmitted to a central processor for the task of acoustic source tracking. Utilizing the characteristics that the speaker state-space model contains linear substructure, Zhang et al [22] combined the auxiliary PF and marginalized PF and developed a distributed marginalized auxiliary particle filter (DMAPF) for acoustic source tracking, which incorporated the current observations into its proposal by introducing an auxiliary variable. An improved distributed unscented Kalman particle filter (DUKPF) is proposed and applied to the single acoustic source tracking problem in noisy and reverberant environments with distributed microphone networks.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have