Guide: Dr. Sanjit Kaul
Team Size: 2
Team Member(s): Anil Sharma, PhD Candidate IIIT-Delhi
Time Period: Aug'14-May'15
Technologies/Concepts Used: Machine Learning, Android, PHP, Matlab, Signal Processing
We investigate an unobtrusive and 24×7 human distress detection and signaling system, Always Alert, that requires the smartphone, and not its human owner, to be on alert. The system leverages the microphone sensor, at least one of which is available on every phone, and assumes the availability of a data network. We propose a novel two-stage supervised learning framework, using support vector machines (SVMs), that executes on a user’s smartphone and monitors natural vocal expressions of fear — screaming and crying in our study — when a human being is in harm’s way. The challenge is to achieve a high distress detection rate while ensuring that the false alarm rate is a manageable overhead, while a typical smartphone user goes about living life as usual. We train the learning framework with carefully selected audio fingerprints of distress and of varied environmental contexts. The audio is used to tune the learning framework to obtain a desirable distress detection rate and false alarm rate (FAR). The ability of the proposed framework to detect distress in rather challenging audio environments is demonstrated.
I worked on this project for a year and coded multiple implementations, on android and matlab. My work also involved testing the algorithm performance, collecting data, testing classifiers and designing the monitoring dashboard.
We demonstrated the impact of environmental context on the detection rate and False Alarm Rate (FAR) trade-off using different point(s) of operation and made the system adaptive to these contexts through environment rejection. We also perform a detailed assesment of the power consumption of the application and optimize the application to run for longer periods through selective classification. Our analysis is based on a user study conducted on 16 volunteers(250hrs data). We monitored the data collection through our dashboard and also identified large-scale and recurring distress patters through clustering and rule-mining.
Our final implementation achieved an accuracy of 93% with an error rate of 1.5%. Exploiting the time contiguous nature of false alarms further allows us to reduce the FAR. We show the feasibility of using our framework anytime and anywhere by testing it over many hours of audio fingerprints recorded by volunteers on their smartphones, as they went about their daily routines. We are able to achieve high distress detection rates at an average overhead that is equivalent to about 1 facebook post every 3 to 4 hours.