O'ZBEK TILI UCHUN NUTQNI TANISH TIZIMINI WAV2VEC2 XLS-R MODELI ASOSIDA ISHLAB CHIQISH VA TURKIY TILLAR O'RTASIDA OVOZDAN MATNGA TARJIMA DASTURIY TA'MINOTI
Keywords:
Kalit so'zlar: avtomatik nutqni tanish, Wav2Vec2 XLS-R, fine-tuning, transfer learning, n-gram til modeli, KenLM, Word Error Rate, turkiy tillar, o'zbek tili.Abstract
Annotatsiya: Ushbu maqolada o'zbek tili uchun avtomatik nutqni tanish (ASR) tizimini yaratishning amaliy yondashuvi taqdim etilgan. Tadqiqot Facebook AI tomonidan taklif etilgan ko'p tilli oldindan o'qitilgan Wav2Vec2 XLS-R modelini transfer learning usuli yordamida o'zbek tiliga moslashtirish (fine-tuning) jarayoniga bag'ishlangan. Ma'lumotlar manbai sifatida Mozilla Common Voice loyihasining o'zbekcha to'plamidan foydalanilgan, model esa Connectionist Temporal Classification (CTC) yo'qotish funksiyasi asosida o'qitilgan. Tanib olish aniqligini oshirish maqsadida, fine-tuning qilingan model KenLM kutubxonasi yordamida qurilgan n-gram til modeli bilan integratsiya qilingan. Shuningdek, o'zbek va turkiy tillar o'rtasida ovozdan matnga tarjimani amalga oshiruvchi dasturiy vositaning IDEF0 metodologiyasi asosida loyihalashtirilgan arxitekturasi va uning Hugging Face platformasiga joylashtirilishi tavsiflangan. Maqolada, shuningdek, turkiy tillar oilasi uchun ASR tizimlarini yaratishda uchraydigan asosiy muammolar — ma'lumotlar tanqisligi, agglyutinativ morfologiya va dialektal o'zgaruvchanlik — mavjud ilmiy adabiyotlar tahlili orqali ko'rib chiqilgan.References
1. Tombaloglu B., Erdem H. Turkish Speech Recognition Techniques and Applications of Recurrent Units (LSTM and GRU) // Journal of Computational Linguistics and Speech Processing. — 2019. — P. 142–167.
2. Tombaloglu B., Erdem H. Deep Learning Based Automatic Speech Recognition for Turkish // Advances in Neural Information Processing. — 2020. — P. 88–103.
3. Mussakhojayeva S., Dauletbek K., Yeshpanov R., Varol H.A. Multilingual Speech Recognition for Turkic Languages // Language Technology and Multilingualism. — 2021. — P. 320–340.
4. Kadyrbek N., Mansurova M., Shomanov A., Makharova G. The Development of a Kazakh Speech Recognition Model Using a Convolutional Neural Network with Fixed Character Level Filters // Journal of Speech and Language Technology. — 2021. — P. 101–118.
5. Taşar D.E., Koruyan K., Çılgın C. Transformer-Based Turkish Automatic Speech Recognition // International Journal of Advanced Computer Science. — 2020. — P. 234–250.
6. Salaeva M., Kuriyozov E., Salaev U. O'zbek tilida chuqur o'rganish texnikalari yordamida avtomatik nutqni tanib olish modellari // Journal of Speech and Language Technology. — 2021. — B. 45–60.
7. Musaev M., Khujayarov I., Ochilov M. O'zbek tilida davomli nutqni tanib olish texnologiyalari // International Journal of Language and Speech Processing. — 2019. — B. 134–149.
8. Mukhamadiyev A., Khujayarov I., Djuraev O., Cho J. End-to-end chuqur neyron tarmoq-hidden Markov modeli va hybrid CTC-e'tibor tarmoqlari // Advances in Speech Recognition Technology. — 2020. — B. 210–230.
9. Salaev U., Kuriyozov E., Matlatipov S., Matlatipov G. Development of a Machine Transliteration Tool for Uzbek Alphabets // International Journal of Central Asian Studies. — 2021.
10. Sharipov M., Salaev U. O'zbek tilida affiksli cheklanmagan holat mashinalari (FSM) yordamida so'zlarni ildiziga ajratish // Central Asian Journal of Linguistics. — 2021. — B. 77–92.
11. Weiss R., Chorowski J., Jaitly N., Wu Y., Chen Z. Introducing Translatotron: An End-to-End Speech-to-Speech Translation Model // Google AI Blog. — 2019, May 15.
12. Wendt C., Lewis W. Evolution of the Microsoft Translator // Microsoft Research Blog. — 2017, December.
13. O'zbekiston Respublikasi Innovatsion rivojlanish vazirligi. Innovatsion nutqni tanish texnologiyalari: MOHIRA AI va MUXLISA AI. — 2021. — mininnovation.uz.
