Phase-locked vocoder

Abstract

The phase vocoder is widely used to provide high-fidelity time stretching or contraction of audio signals such as speech or monophonic musical passages. Two problems bedevil the reconstructive phase of this technique. First, the frequency estimate is usually multi-valued and one does not know how to choose which of the possible frequencies given by the analysis to use in resynthesis. Second, a sinusoidal component in the incoming sound always excites several channels of the analysis; the reconstructed sine waves can interfere with each other, giving rise to a reverberant sound. An improved formulation of the phase vocoder is proposed for which the first difficulty does not arise; and a means of phase-locking adjacent channels of the resynthesis is proposed which alleviates the second one.