Engineering Commons^™

Uzbek Commands Recognition By Processing The Spectrogram Image, M M. Musayev, I Sh Khujayorov, M I. Abdullaeva, M M. Ochilov

Technical science and innovation

This paper describes the most common algorithms with image approach convolutional neural network and two-dimensional DCT with machine learning classification KNN, SVM and RF. These algorithms are evaluated for applicability to the Uzbek language and a comparative analysis on the accuracy and recognition rate. The command words of the Uzbek language were chosen for the experiments. According to the results, it was found that both methods give high rates of recognition accuracy and are 92% (CNN) and 90% (2DDCT+Zigzag+SVM). Also the combinations of 2D-DCT+Zigzag+ KNN and 2D-DCT+Zigzag+ RF with average recognition accuracy of 86% and 85%, respectively, were considered in …