Experimental Phonetic Research Interlingual Interference and Accent in the Russian Speech of Native Speakers of the Kabardino-Circassian Language

The use of deep learning algorithms has made it possible to achieve a human parity in telephone conversational speech recognition. However, separate tasks, among which the problems of noise reduction, automatic segmentation of a mixed audio signal, and non-native speech recognition, remain open. The complexity of the accent speech recognition is the consequence of the discrepancy between the non-native speech and native language resources used in training, acoustic, language and pronunciation modeling based on data-driven approach. The existing approaches to solving these difficulties based on the use of non-native speech resources or multilingual corpora are ineffective. It is necessary to study the specifics of the psychophysiological mechanisms of bilinguals who use two or more language systems in communication. This paper shows the results of the experimental phonetic study of interlingual interference and accent in the Russian speech of native speakers of the Kabardino-Circassian language. The preliminary acoustic analysis of the initial formant measurements of non-native Russian speech of Kabardino-Circassian speakers at the phonetic level showed a systematic deviation of the acoustic patterns of vowel allophones (quantitative estimates of the vowel allophone [A] are given for brevity) in the F1-F2 space. The aimed is to develop methods for modeling the non-native speech for subsequent use in speech recognition systems, language identification and accents in conditions of limited linguistic resources. The obtained experimental data can be useful for the development of the theory of contrastive acoustic analysis of vocal systems of languages with different structures, typical interfering pronunciation errors detection, and also for the development of theoretical models of foreign languages learning.