Estimation and tracking of fundamental, 2nd and 3d harmonic frequencies for spectrogram normalization in speech recognition

A stable and accurate estimation of the fundamental frequency (pitch, F0) is an important requirement in speech and music signal analysis, in tasks like automatic speech recognition and extraction of target signal in noisy environment. In this paper, we propose a pitch-related spectrogram normalization scheme to improve the speaker – independency of standard speech features. A very accurate estimation of the fundamental frequency is a must. Hence, we develop a non-parametric recursive estimation method of F0 and its 2nd and 3d harmonic frequencies in noisy circumstances. The proposed method is different from typical Kalman and particle filter methods in the way that no particular sum of sinusoidal model is used. Also we tend to estimate F0 and its lower harmonics by using novel likelihood function. Through experiments under various noise levels, the proposed method is proved to be more accurate than other conventional methods. The spectrogram normalization scheme makes a mapping of real harmonic structure to a normalized structure. Results obtained for voiced phonemes show an increase in stability of the standard speech features – the average within-phoneme distance of the MFCC features for voiced phonemes can be decreased by several percent.