Vocal tract resonances tracking by auxiliary particle filters

This paper presents an auxiliary particle filtering method for vocal tract resonances (VTRs) tracking. The proposed method is based on a version of the structured speech model, with a state-space formulation consisting of target-guided dynamics and a non-linear prediction mapping from resonance frequencies and bandwidths to linear prediction coding cepstra (LPCC). In conventional particle filter, the particles are generated from the transition prior can easily land on low-likelihood areas thus wasted. To overcome this difficulty, we explore a novel algorithm to generate sophisticated proposal distribution that integrates the phoneme information by introducing an auxiliary vector. Experimental results show that due to the improvement of proposal distribution, the proposed method is able to track the VTRs efficiently using a small number of particles both in the vocalic regions and the consonantal regions of the speech utterance.