Speech Signal Structuring Method for Biometric Personality Identification

摘要

This paper proposes a method for structuring a speech signal. For this, segmentation method, methods for determining the fundamental tone of the vocal segment and determining on its basis the boundaries of the quasiperiodic oscillations of the vocal segment, the geometric transformation of quasiperiodic oscillations of the vocal segment were suggested. The proposed segmentation of the speech signal uses statistical estimation of short-term energies, which allows the use of an adaptive threshold, thus increasing the vocal segments determination accuracy. The proposed definition of fundamental tone of the vocal segment uses bandpass filtering and statistical estimation of local extremum, which reduces computational complexity, and also reduces noise dependency and allows the use of an adaptive threshold, thus increasing the accuracy of determining the fundamental tone and the boundaries of quasiperiodic oscillations of the vocal segment. The proposed geometric transformation of quasiperiodic oscillations of the vocal segment allows you to transform quasiperiodic oscillations to a single amplitude-time window, which allows you to form frames of the vocal segment, taking into account its structure.