Ai-based offline speech recognition for kazakh, russian and english languages

This paper presents a novel full autonomous multilingual audio transcription system tailored to Kazakh, Russian, and English. The proposed solution integrates a language detection module based on SpeechBrain with a transcription engine using Vosk, and employs FFmpeg for robust audio preprocessing. The system automatically detects the language from the initial 10 seconds of an audio stream, selects the corresponding acoustic model, and produces an accurate text transcription. Experimental evaluations on both synthetic and real audio data indicate that our approach achieves competitive performance in terms of accuracy (with word error rates ranging from 5 to 10% under optimal conditions) and processing speed, while operating entirely on local resources without dependency on cloud services. These features make it particularly suitable for applications in digital forensics and other domains that require secure real-time transcription capabilities.