Localization and tracking of acoustic sources in room environment

• Main Author: Wu, Kai
• Other Authors: Andy Khong Wai Hoong
• Format: Theses and Dissertations
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: http://hdl.handle.net/10356/72519
• Institution: Nanyang Technological University

This thesis addresses two areas of the acoustic source localization and tracking (ASLT) problem in a room environment, namely, tracking of acoustic sources using multiple omni-directional microphone arrays and DOA estimation of the acoustic sources using a single acoustic vector sensor (AVS). The challenges for the ASLT problem, in both aforementioned applications may include room reverberation, background environmental noise, sound interference, as well as the presence of multiple speakers. For multiple omni-directional arrays, the thesis focuses on the source tracking problem where the source position is estimated sequentially across several time frames using the particle filter (PF) framework. To achieve single-source tracking in a reverberant and noisy environment, an algorithm which utilizes the well-known sequential importance resampling PF (SIRPF) framework is proposed. As will be shown in this thesis, this proposed algorithm derives the measurement likelihood which is robust to reverberation and noise. For single-source tracking in the presence of sound interference, another SIRPF-based algorithm is proposed. This algorithm exploits the harmonicity feature of a speech signal for deriving the measurement likelihood. Due to the use of distinctive speech feature, speech-sensitive tracking can be achieved in the presence of sound interference. The performance of these two algorithms have been verified through simulation. The problem of tracking of alternating speakers will then be discussed in which the speech sources are active in turns. For solving this problem, a novel swarm intelligence based PF (SWIPF) which jointly exploits the advantages of PF and particle swarm intelligence is proposed. The PF framework is used as sequential state estimation framework which suits for the tracking problem. The limitation of PF, which lies in the particle sampling, is addressed by incorporating the particle swarm intelligence. By using the swarm intelligence, particles are associated with interaction and memory mechanisms. When alternation occurs, particles can be directed toward the true source location by interacting and sharing the fitness information among themselves. In addition, the memory mechanism allows particles to retain their previous best-fit positions when signals are corrupted by noise and reverberation. The proposed SWIPF is verified using both simulations and real experiments. The thesis finally considers the multi-source DOA estimation problem using an AVS. Unlike the conventional microphone arrays which requires inter-spacing between microphones, the co-location of sensor elements in an AVS can be exploited to achieve robust DOA estimation in a reverberant environment. As opposed to the existing multi-source DOA estimation algorithms using AVS, the proposed algorithm is developed from a reverberant received signal model. By exploiting the co-location of the sensor elements in an AVS, the low-reverberant-single-source (LRSS) zones of the received signals, where only one source is dominant with a high signal-to-reverberation ratio, can be identified. By using only these identified LRSS zones followed by a clustering step, multi-source DOA estimation in reverberant environment can therefore be achieved. Simulation is conducted to verify the performance of the proposed algorithm.