IMPORTANCE SAMPLING PARTICLE FILTER FOR ROBUST ACOUSTIC SOURCE LOCALISATION AND TRACKING IN REVERBERANT ENVIRONMENTS

The concept of acoustic source localisation and tracking (ASLT) plays an important role in many practical speech acquisition systems. Exact knowledge of the speaker position is usually the key to acquiring clean speech using e.g. beamforming or equalisation. Multipath sound propagation in practical environments however constitutes a major challenge to overcome for any array-based tracker. The performance of methods used traditionally for this purpose, such as steered beamforming (SBF) and time delay estimation (TDE), are heavily influenced by reverberation and background noise. Recently, the concept of particle filtering (PF) was proposed as a new approach to this problem [1, 2, 3]. This method relies on a Bayesian filtering principle which can be summarised as follows.