Abstract
Recent breakthroughs in artificial intelligence and deep learning have fundamentally transformed the landscape of spoken language processing technologies. Automatic speech recognition (ASR) and text-to-speech (TTS) synthesis have emerged as essential components driving digital accessibility across diverse linguistic communities. The Sakha language, representing the northeastern branch of the Turkic language family, continues to face substantial technological barriers stemming from insufficient digital resources, limited annotated corpora, and the absence of production-ready speech processing systems. This comprehensive investigation examines the feasibility and effectiveness of adapting contemporary transformer-based neural architectures for bidirectional speech conversion tasks in Sakha. Our research encompasses detailed analysis of encoder-decoder frameworks, specifically OpenAI’s Whisper large-v3 and Meta’s Wav2Vec2-BERT for voice-to-text transformation, alongside Coqui’s XTTS-v2 system for text-to-voice generation. Particular emphasis is placed on addressing linguistic and technical obstacles inherent to Sakha, including its complex agglutinative morphological structure, systematic vowel harmony patterns, and distinctive phonemic inventory featuring sounds absent from most Indo-European languages. Experimental evaluation demonstrates that comprehensive fine-tuning of Whisper-large-v3 achieves exceptional recognition accuracy with word error rate (WER) of 8%, while the self-supervised Wav2Vec2-BERT architecture attains 13% WER when augmented with statistical n-gram language modeling. The neural synthesis system exhibits robust performance despite minimal training data availability, achieving average loss of 2.49 following extended training optimization and practical deployment via Telegram messaging bot. Additionally, ensemble meta-stacking combining both recognition architectures achieves 27% WER, demonstrating effective complementarity through learned hypothesis arbitration. These findings validate transfer learning methodologies as viable pathways for developing speech technologies serving digitally underrepresented linguistic communities.