MULTILINGUAL NEURAL MACHINE TRANSLATION (NMT) MODEL FOR ADAPTING KAZAKH, RUSSIAN, AND ENGLISH SPEECH TO PROTOCOL FORMAT

Abstract

This study comprehensively addresses the problem of automatically translating speech in Kazakh, Russian, and English and adapting it to an official protocol format. The work provides an in-depth analysis of the scientific foundations of multilingual neural machine translation (NMT) models based on the transformer architecture and demonstrates methods for integrating them with automatic speech recognition (ASR) systems. The proposed approach encompasses the full processing pipeline from audio signals to structured official documents, including transcription, translation, post-processing, and formatting text into an official style. The results show that multilingual transformer models are capable of producing high-quality translations among Kazakh, Russian, and English. Additionally, the method enables automatic adaptation of text to a protocol format by removing disfluencies, restoring punctuation, maintaining terminological consistency, and normalizing text to an official style. The proposed approach significantly accelerates the documentation of multilingual meetings, facilitates information structuring, and contributes to the development of Kazakh digital language resources.