Vocal tract resonances tracking by auxiliary vector particle filters

An auxiliary vector particle filtering method was presented for vocal tract resonances(VTRs) tracking. The proposed method uses particle filter based on a version of state-space model which describes the characteristics of speech signal.The speech model consists of a target-guided dynamic function and a non-linear prediction mapping from resonance frequencies and bandwidths to LPC cepstra(LPCC).There are two characteristics in the proposed method. First,particle filtering technique is proposed to solve the non-linear problem of speech model.Second,an auxiliary vector,embedded in the state function of speech model,is used to incorporate the most current observations and to generate the proposal distribution of particle filter.Experimental results show that the proposed method is able to track the VTRs of fluent speech utterance efficiently with a small number of particles and able to solve the problem of spurious peaks and merging peaks.