Joint time delay and pitch estimation for speaker localization

Abstract

In this paper, we attempt to develop an efficient and accurate algorithm for joint time delay and pitch estimation of a speech signal received at a microphone array. The time delay measurement allows a speaker to be located while the detection of the pitch frequency is useful for analyzing the acoustic properties of the sound. A subspace method based on state-space realization is first introduced for joint time delay and frequency estimation of a synthetic signal consisting of several frequency components. The frequency estimates are obtained directly from the eigenvalues of the state transition matrix whilst the time delay is approximated from the observation matrix and the estimated frequencies using a least square approach. The method is then extended to track both the time delay and pitch frequency of a speech signal, which is modeled by a summation of sinusoids that are harmonically related to the fundamental frequency (pitch) and spectrally shaped by the vocal tract transfer function. Extensive simulation tests have been done to validate the effectiveness and accuracy of the proposed algorithm.