Sound based location detection using machine learning methods

Today, sound plays a significant role in every aspect of human life. The number of crimes committed in every environment, from personal security to critical inspections in our environment, is increasing. With this increase, the identification of criminals and the clarification of events are of great importance. Audio data can be used as descriptive information to track the criminal. Activities carried out in our homes produce different sound signals. Various activities are carried out in other locations in the house, and the acoustic properties of the sound in each location differ. As a result of the detection of the produced sound signals, the person's position can be estimated. For this purpose, a new sound dataset consisting of household sound was collected. Videos related to activities to be performed according to home locations on YouTube were used in the generated dataset.In this thesis, the popular machine learning classifiers that have been used by recent research have been applied, such as KNN, SVM, random forest (REF), etc Finally, our model, the Cubic-SVM, determined the position of the classified sound and achieved a good result with an accuracy of 96.85 for identifying the location of the classified sound, compared with existing results obtained by the researchers.