Comparison of Pitch Tracking Methods for Speech in Additive White Gaussian Noise
David A. Krubsack · e-Publications@Marquette (Marquette University) · 1986
An exploration of five methods (and variations of these methods) for the estimation of the fundamental frequency of a voiced speech signal corrupted by high levels of additive white Gaussian noise is discussed. The time domain methods explored include the autocorrelation and the cepstrum. The frequency domain methods explored include the evaluation of the magnitude Fourier transform, the autocorrelation of the magnitude Fourier transform, and the harmonic sum spectrum. All methods are evaluated on a fixed window basis and quantitative results are presented for seven signal-to-noise ratios. In addition, this thesis includes a discussion of several issues related to pitch tracking as well as some general conclusions regarding the pitch tracking methods explored.