Combining multiband joint position-pitch algorithm and particle filters for speaker localization

We present a combination of the multiband joint position-pitch (M-PoPi) estimation algorithm with the particle filtering framework to enhance the localization accuracy when tracking multiple concurrent speakers. A new likelihood function derived from the M-PoPi algorithm is proposed for the particle filter framework. The performance of the particle filter based tracker is compared with the M-PoPi algorithm. The proposed framework improves localization accuracy for all cases ranging from single upto three concurrent speakers.

[1]  Eric A. Lehmann,et al.  Particle Filter with Integrated Voice Activity Detection for Acoustic Source Tracking , 2007, EURASIP J. Adv. Signal Process..

[2]  M. Kepesi,et al.  Joint Position-Pitch Estimation for Multiple Speaker Scenarios , 2008, 2008 Hands-Free Speech Communication and Microphone Arrays.

[3]  Jon Barker,et al.  Using location cues to track speaker changes from mobile, binaural microphones , 2009, INTERSPEECH.

[4]  Fabrice Plante,et al.  A pitch extraction reference database , 1995, EUROSPEECH.

[5]  Darren B. Ward,et al.  Particle filtering algorithms for tracking an acoustic source in a reverberant environment , 2003, IEEE Trans. Speech Audio Process..

[6]  Franz Pernkopf,et al.  Joint Position-Pitch Tracking for 2-Channel Audio , 2007, 2007 International Workshop on Content-Based Multimedia Indexing.