Automatic estimation of experience of privacy using acoustic cues

Privacy is a fundamental aspect of human interactions and with the growing popularity of tracking and recording devices, we are becoming more aware of its significance. If the perceived level of privacy can be analyzed automatically by the electronic devices, information sharing could be regulated according to the privacy requirements of the users. This work aims to automate the process of estimation of the perceived levels of privacy using acoustic cues using recent advancements in the field of machine learning and deep learning. Previous work related to acoustic sound classification has been discussed in the thesis and machine learning models- namely random forests and convolution neural networks to automatically evaluate the level of perceived privacy using acoustic cues. The results indicate that it is possible to automatically evaluate the level of privacy using acoustic cues to train machine learning models. A validation accuracy greater than 60% was obtained through deep learning Convolution Neural Network models. There is a high variability of results due to cultural factors and individual bias and this can be explored in future work