Microsoft Word Tezis-Salayeva-ict
Download 38.26 Kb. Pdf ko'rish
|
Tezis-Salayeva-ICT
- Bu sahifa navigatsiya:
- , Ulugbek Salaev 1
1 TATU, 2023 Creating Speech Recognition Models for Uzbek Language Makhliyo Salaeva 1 , Elmurod Kuriyozov 1,2 , Ulugbek Salaev 1 1 Urgench State University 2 Universidade da Coruna, Spain E-mail: makhliyo.salaeva@urdu.uz* , elmurod1202@urdu.uz , ulugbek.salaev@urdu.uz Abstract This paper presents a study on the use of automatic speech recognition (ASR) models for the Uzbek language. The performance of pre-existing ASR models for Uzbek is evaluated and compared, as well as the methods for creating ASR models for low- resource Uzbek language. The study includes the use of transfer learning and data augmentation techniques to improve the performance of ASR models on low-resource languages. Additionally, the effectiveness of unsupervised learning algorithms for training ASR models on low-resource languages is also discussed. The results of this study have important implications for the development of ASR technology for under-resourced languages. Introduction Automatic speech recognition (ASR) is a technology that allows computers to transcribe and translate spoken language into text. It has a wide range of applications, from voice-controlled assistants to speech-to-text dictation software. However, the performance of ASR models can vary greatly depending on the language being recognized. For languages with ample resources, such as English, ASR models have achieved high accuracy levels. However, for low-resource languages, such as Uzbek, the development of accurate ASR models is a challenging task. The Uzbek language is the official language of Uzbekistan, spoken by more than 30 million people worldwide. Despite its wide usage, the resources for training ASR models for Uzbek are limited. This research aims to analyze the current state of ASR models for Uzbek and investigate ways to improve the performance of ASR models for low-resource languages such as Uzbek. The study will focus on comparing the performance of pre- existing ASR models for Uzbek and evaluate the effectiveness of various methods for creating ASR models for low-resource languages, including transfer learning and data augmentation techniques. Additionally, the study will also investigate the effectiveness of unsupervised learning algorithms for training ASR models on low-resource languages. The results of this research will have important implications for the development of ASR technology for under-resourced languages such as Uzbek. It will provide insights into the challenges and limitations of developing ASR models for low-resource languages and offer potential solutions to improve the performance of ASR models in such scenarios. There are a few different approaches you could take when analyzing ASR models for the Uzbek language, depending on the specific focus of your research. One option would be to compare the performance of different pre-existing ASR models for Uzbek, such as those developed by Google, Microsoft, or other companies, and evaluate their accuracy and speed in recognizing Uzbek speech. Another option would be to train your own ASR model using a dataset of Uzbek speech, and compare its performance to the pre-existing models. In terms of creating models for low-resource Uzbek language, one approach would be to use transfer learning techniques to fine-tune a pre-trained model on a smaller dataset of Uzbek speech. This can help to improve the model's performance on low- resource languages. Moreover, data augmentation techniques such as adding background noise or varying the speed of the speech can also be used to increase the size of the dataset and improve the model's performance. 2 Furthermore, you could also look into the use of unsupervised learning algorithms for training ASR models on low-resource languages, which do not require large amounts of labeled training data and can be more effective in these scenarios. Download 38.26 Kb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling