Monitoring of the natural voice variations in open and closed phases with frequency warped ARMA modeling

Abstract

The objective of this paper is to propose the use of a speech model with psychoacoustical information to distinguish between the open and closed phases of the vocal folds, in order to monitor formants as phonetic speech characteristics. Taking a frequency warped ARMA model for each one of the phases, the aim is to integrate the information of various consecutive periods in which the poles and the zeros that form the vocal tract can be considered common due to its slow variation. To analyze the capacity of phonetic monitoring, different phonetic voice transitions of speech registers are used from a database that provides information on glottal closings. First, we show the dependency with the warping factor. Then, the consistency and reliability of the method is demonstrated, as well as its better performance compared to no warped ones. And finally, we will see the well behavior against moderate noise.