The method of real-time acoustic measurement of dynamical changes in the speaker’s emotional state

Abstract

In this paper we consider the issues in implementations of interactive voice response systems with remote access. Their efficiency can be improved by automatically analyzing changes in the user's emotional state during the dialogue. In order to measure the indicator of the dynamics of the emotional statein real time, it is proposed to use the effect of sound (phonetic) variability of the user's speech at short intervals (fractions of a minute). The novel method of acoustic measurements in conditions of small samples has been developed based on information-theoretic approach by using a scale-invariant gain-optimized dissimilarity measure of the speech signals in the frequency domain. An example of its practical implementation in soft real time is considered. It is shown that the delay in obtaining the measurement results does not exceed in this case 10–20 sec. The experimental results confirmed the high speed of the proposed method and its sensitivity to changes in the emotional state under the influence of external noise. The proposed method can be used for automated quality control of voice samples of users in unified biometric systems, as well as to improve safety by non-contact identification of potentially dangerous persons with short-term psycho-emotional disorders.