Sequential Monte Carlo Fusion of Sound and Vision for Speaker Tracking