Kapalı ortama uygun aktivite tespiti için derin öğrenme ile ses analizi

Smart systems analyze acquired data from the environment and generate results that make the user's life easier. Smart systems improve the quality of life for users by analyzing data collected from the environment. The scope of application and demand for smart systems expands daily. The success of smart systems depends on the quality and quantity of the data used for analysis. For this reason, sound data is becoming increasingly popular as a result of its accessibility and a vast number of samples. In our work, we aim to build a smart system that is affordable, mobile, and easily accessible. We transformed sound data into spectrograms and used them to train a convolutional neural network. Three distinct approaches (image combination, multi-input, and transfer learning) were used during training. The highest success rate is achieved by the transfer learning approach with 87.26 percent. From the ESC 50 dataset, eight sounds are selected that are suitable for indoor environments. In order to assess the trained models' performance in the real world, we also use recordings of real-life examples. Even though the overall results are not high as the ESC dataset test, the system achieved almost 100 percent for some sound types. For this purpose, a new smartphone prototype application is developed to record and analyze sound.