An Algorithm for Formant Estimation of Whispered Speech

Whispered speech, always in low SRN, is more difficult in its formant estimation. This paper proposes an algorithm for formant estimation of whispered speech. It includes three subroutines: calculation of the autocorrelation function (ACF) of whispered speech, segmentation of the ACF spectrum to get the coefficients of the inverse-filter, inverse-filter control (IFC) of the ACF spectrum and calculation of the formant frequencies. The tests are carried on Chinese whispered vowels, and the proposed algorithm is proved to be efficient. Especially in low SNR, it is superior to conventional methods for formant estimation of whispered speech