<![CDATA[Sound detection and keyword recognition in audio signals have become rapidly growing research areas due to their wide range of applications, from intelligent audio surveillance to human-computer interaction systems. This study aims to develop a deep learning model based on Convolutional Neural Networks (CNN) to automatically detect and classify specific words in speech recordings. The focus of this research is the detection of the keywords "dog" and "children" contained in speech data. The research methodology includes data preprocessing through noise reduction and normalization, as well as data augmentation techniques such as pitch shifting to improve the robustness of the model. Audio features are extracted using the Short-Time Fourier Transform (STFT) to generate visual representations in the form of spectrograms, which serve as the primary input to the CNN architecture. Experimental results show that the developed model successfully classified the target keywords with an accuracy of 90,00%. The model proved effective in recognizing both spectral and temporal patterns of spoken keywords and has the potential to be implemented in real-time sound detection systems.]]>
Copyrights © 2026
