Speech modeling with time varying amplitude and phase

Formant transitions contain important information related to hearing and perception. The concept of moving poles is used in this paper in relation to the time-varying impulse response model of the vocal tract and describes the trajectory of a formant. The model allows for amplitude modulation as well as other nonstationarities present such as formant bandwidth variations. The moving poles are modeled as polynomial functions of time, whose linear (albeit suboptimal) estimates ate obtained from the higher-order ambiguity function. An overall optimal but nonlinear least squares scheme is developed to improve upon the linear estimates and both algorithms are tested by simulations