This paper introduced a new speech production model aiming

at synthesizing natural speech in real-time by modeling

the key dynamic properties of the articulators in a nonlinear

state-space framework. The goal-oriented movement of the

tongue tip, tongue dorsum, upper lip, lower lip and jaw are

described in a linear state equation. The so produced articulatory

trajectories combined with the effects of velum and

larynx are mapped into acoustic features in the nonlinear

observation equation. The input and output of the model are

time-aligned phone sequence and speech waveform respectively.

This speech production model can also be directly

applied to speech recognition to better account for coarticulation

and phonetic reduction phenomenon with considerably

less parameters than the traditional HMM based approaches.

PDF file

In  Proc. of the Int. Conf. on Acoustics, Speech, and Signal Processing

Details