An algorithm for formant frequency estimation from noise-corrupted speech signals

Abstract

An algorithm for format frequency estimation from noise-corrupted speech signals is presented. A scheme for frequency domain noise reduction and the ACF of the resulting noise-compensated speech signal is then employed in a modified form of least-squares Yule-Walker equations (LSYWE) to estimate poles of the VT system. A pole selection criterion is estimated for extracting the desired formats that enables the proposed scheme to avoid errors associated with weak formats. The proposed formant frequency estimation algorithm has been tested using different synthetic vowels synthesized using the Klatt synthesizer and natural vowels from the North Texas speech database as well as some natural sentences from the TIMIT speech database. Experimental results on natural and synthetic speech signals show the efficacy of the proposed method in estimating formant frequencies at a moderate to low levels of SNR.