Method of real-time updating for voice templates in the Unified Biometric System

Abstract

The article was written in the development of ideas from a previous work of the authors [5]. The task of automated quality control of voice templates, which were registered and stored in the Unified Biometric System is considered. A solution to the problem of timely updating of the collected templates is proposed, since over time they lose their consumer qualities. A new indicator of the acoustic quality of voice templates in the Kullback–Leibler information metric was investigated and a method for measuring it at the moments when users contacting the system with service requests was proposed. An example of the practical implementation of the proposed method is shown. Using the author's software, a full-scale experiment was conducted, quantitative estimates of the period for updating voice templates were obtained, and recommendations were given on their practical application. The results can be used to develop new and modernize existing systems and technologies for automated quality control and updating of biometric personal data templates.

A Method of Real-Time Dynamic Measurement of a Speaker’s Emotional State from a Speech Waveform

“…Due to the rapid development of information-communication technologies dialogue systems with a voice interface for remote accommodation of the population are being widely disseminated throughout the world. [1,2]. Their effectiveness depends to a signifi cant degree on the dynamics of the emotional state of users (DESU) of information systems and timely detection of changes of this state during the course of the dialogue in real time [3,4].…”

“…Meanwhile, the acuteness of the problem of high speed of operation of automatic speech processing can be greatly ameliorated by transition to autonomous (without Internet access) algorithms with tuning to the voice of each specifi c user [20][21][22]. The principle of operation of such algorithms is based on phonetic analysis of speech with sequential partitioning of a speech waveform into frames of short (τ = 10-15 ms) duration not exceeding the limits of the duration of a minimum speech unit (MSU) such as phonemes and their allophanes in the oral speech of the speaker [2,12]. In them, the role of voluminous (tens and hundreds of gigabytes) speech databases are fi lled by personal phonetic databases that are generated at the stage of data preparation by a known procedure [2,3].…”

“…The principle of operation of such algorithms is based on phonetic analysis of speech with sequential partitioning of a speech waveform into frames of short (τ = 10-15 ms) duration not exceeding the limits of the duration of a minimum speech unit (MSU) such as phonemes and their allophanes in the oral speech of the speaker [2,12]. In them, the role of voluminous (tens and hundreds of gigabytes) speech databases are fi lled by personal phonetic databases that are generated at the stage of data preparation by a known procedure [2,3]. This article is devoted to the study in this timely area for acoustic measurements of the index of DESU.…”

“…of the mean deviation of the instantiations of ρ(x r, j |x r * ) ≜ ρ r (x r, j ) of a relatively informational central-standard x r * ; where ≜ is the symbol for equality by defi nition [2,25,26].…”

“…3) and the set of alternative spectral densities of power {G r ( f )} from the output of the unit 11 for recording and storage of phonetic references is used for this purpose. The content of this unit is generated by the observer by the known procedure [2,25] at the data preparation stage in the task of DESU measurements. From the results of recognition, the appropriate cluster X r is instantaneous replenished with a new component.…”