O‘ZBEK TILI SHEVALARINI AVTOMATIK NUTQNI ANIQLASH (ASR) TIZIMLARIDA TANIB OLISH SAMARADORLIGI TAHLILI
This article comparatively investigates the impact of Uzbek dialectal diversity on the performance of automatic speech recognition (ASR) systems. The relevance of the study lies in the performance degradation of intelligent models during live dialectal speech processing. Through a statistical analysis of audio samples from the Karluk, Kipchak, and Oghuz dialect groups, the capabilities of Kotib STT, OmoN STT, Rubai STT, and Whisper Large models were evaluated. Results indicate that while recognition accuracy for the standard literary language reaches 90–95%, it drops to 25–30% for highly variable dialects, primarily due to lexical errors caused by affixal reduction and vowel alternation. To mitigate this issue, the study scientifically justifies the need to expand multi-dialectal acoustic-linguistic corpora and fine-tune models based on regional linguistic characteristics.
1. Radford A., Kim J. W., Xu T. et al. Robust Speech Recognition via Large-Scale Weak Supervision International Conference on Machine Learning. - PMLR, 2023. - P. 28492-28518.
2. Robinson N. R., Sun K., Xiao C. et al. JHU IWSLT 2024 Dialectal and Low-resource System Description IWSLT Conference Proceedings. – 2024.
3. Elmahdy M., Gruhn R., Minker W. Novel Techniques for Dialectal Arabic Speech Recognition. – Springer, 2022. – 180 p.
4. Agarwal A., Zesch T. Robustness of end-to-end Automatic Speech Recognition Models – A Case Study using Mozilla DeepSpeech proceedings of the 17th conference on natural language processing (konvens 2021)
5. Hamroyeva Sh. M., Maxmudjonova G. U. Kam resursli tillarda g2p model tayyorlashda fonetik korpuslarning ahamiyati: o‘zbek tili misolida “Kompyuter lingvistikasi: muammolar, yechim, istiqbollar” v xalqaro ilmiy-amaliy konferensiya 2025
6. Po'latov A. Kompьyuter lingvistikasi. – Toshkent: Akademnashr, 2023. – 288 b.
7. O‘zbek tili milliy korpusi va nutqni matnga aylantirish (ASR) muammolari O‘zbekistonda sun’iy intellekt va raqamli transformatsiya konferensiyasi materiallari. – Toshkent, 2025. – B. 88-96.
8. Salaeva M., Kuriyozov E., Salaev U. Uzbek automatic speech recognition models using deep learning techniques "Kompyuter lingvistikasi: muammolar, yechim, istiqbollar" xalqaro ilmiy-amaliy konferensiyasi materiallari. – Toshkent, 2023. – Vol. 1. – № 01. – B. 218-223.
9. Alayev R. H., Elov B. B., Hamdamov O`. Maʼlumotlar to`plamini o`qitish, baholash va test to`plamlariga ajratish usullari. 2024.
10. Speech-to-Text Models in Uzbek Language: Achievements and Limitations Fayzullo Nazarov; Akbar Soliev; Bunyod Eshtemirov 2026
Copyright (c) 2026 «ACTA NUUz»
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
.jpg)
1.png)
