ACOUSTIC INTERFERENCE IN EMOTIONALLY COLORED SPEECH: AI MODEL-ING OF GERMAN PAREMIAS (BASED ON THE SPEECH OF AZERBAIJANI-GERMAN BILINGUALS)

Abstract This article investigates the phenomenon of acoustic interference in the context of Azerbaijani-German bilingualism during the articulation of emotionally colored formulas, specifically traditional blessings and curses. The study focuses on the specific realization of vocalism and consonantism, as well as the critical challenge of decoding such speech using modern Artificial Intelligence (ASR – Automatic Speech Recognition) systems. Rooted in L.R. Zinder’s phonological theory of the independent phoneme and contemporary research in cultural pragmatics, the study demonstrates that the cognitive base of a bilingual speaker—operating simultaneously within two distinct linguistic codes—generates highly stable interference patterns under affective stress. The experimental phonetic analysis reveals that when confronting complex German consonant clusters in expressive vocabulary, Azerbaijani native speakers unconsciously resort to vowel epenthesis and distort vowel formants (e.g., the open monophthong [ɛː]). While the human cognitive system easily decodes these pragmatic deviations, artificial neural networks trained on monolingual datasets interpret such phonetic noise as structural failures, leading to significant Word Error Rates (WER) and the complete collapse of lexical recognition. The findings emphasize that culturally determined speech acts performed by bilinguals require the development of new, inclusive, and pragmatically annotated training datasets to enhance the adaptability and accuracy of future AI models.