Year: 2025
Type: article
Abstract: This thesis presents the fundamental methods of automatic speech-to-text (ASR) technologies. Firstly, the Dynamic Time Warping (DTW) method, which identifies similarities between signals, is described. Then, Hidden Markov Models (HMM) and their application in modeling probabilistic sequences of states are explained. The role of Gaussian Mixture Models (GMM) in acoustic modeling and their function in matching speech features to phonemes are highlighted. The advantages of neural networks in speech recognition, including their ability to process in parallel and generalize patterns, are discussed. Furthermore, modern approaches based on Deep Neural Networks (DNNs) and Recurrent Neural Networks (RNNs) are shown to improve the quality of speech recognition. Language models and adaptation methods are also reviewed. These approaches are of significant importance in improving speech recognition systems for the Uzbek language.
Source: Neliti
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7132734500