Pitch Tracking of Acoustic Signals based on Average Squared Mean Difference Function

TL;DR

This paper introduces a novel pitch tracking method that minimizes local variance of periodically sampled acoustic signals to accurately estimate fundamental frequency across different speakers.

Contribution

The paper presents a new variance minimization approach for pitch tracking using average squared mean difference function, applicable to text-independent voiced signals.

Findings

Effective pitch estimation across multiple speakers

Robust to variations in voiced signals

Improves accuracy over existing methods

Abstract

In this paper, a method of pitch tracking based on variance minimization of locally periodic subsamples of an acoustic signal is presented. Replicates along the length of the periodically sampled data of the signal vector are taken and locally averaged sample variances are minimized to estimate the fundamental frequency. Using this method, pitch tracking of any text independent voiced signal is possible for different speakers.