Microsoft Word Tezis-Salayeva-ict


Download 38.26 Kb.
Pdf ko'rish
bet1/3
Sana18.06.2023
Hajmi38.26 Kb.
#1580780
  1   2   3
Bog'liq
Tezis-Salayeva-ICT



1
TATU, 2023
Creating Speech Recognition Models for Uzbek Language
 
 
Makhliyo Salaeva
1
Elmurod Kuriyozov
1,2
, Ulugbek Salaev
1

Urgench State University 

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:
  1   2   3




Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling