Comparison of different filter approaches for the online frequency analysis of speech signals
Andreas Rauh, Susann Tiede, Cornelia Klenke · 2017
The fundamental building block of spoken languages is a list of phonemes from which syllables and, hence, also words are formed. A systematic distinction between these phonemes becomes possible by the characteristic frequency components that are included in each sound. On the one hand, voiced phonemes are characterized by several sharp frequency components. On the other hand, wide, typically blurred frequency spectra are common for unvoiced sounds. Software-based assistance systems for the automatic classification of phonemes, therefore, have to estimate the variation of frequencies and their associated bandwidths that are included in a speech signal. This paper gives a comparison of different stochastic filtering approaches for the online estimation of the formant frequencies of phonemes from both a methodological and an application-oriented point of view. This approach for online frequency estimation is an integral component for the development of the assistance system SUSE (A Software assistance system for Uncovering speech disorders by Stochastic Estimation techniques) which aims at supporting the detection and classification of linguistic disorders in the everyday work of speech therapists.