The government mandated an SOS or panic button on all mobile phones sold in India after the horrific Nirbhaya gang rape incident. While the Indian handset makers have put an actual button on the phones (especially feature phones), international brands have included an app that allows users to download a soft key on the screen.
Now in a real situation, none of these are of much use. If a woman is attacked, will she grope for the phone in her bag, find the key and press – or fend off the attack? That is why the Rakshak app is a real innovation. It requires no action on the part of the victim. The phone is “listening”, when the app is on. If the audio microphone hears words such as “help” or “bachao”, it uses the machine learning algorithm to judge the emotional state from sound, pitch, etc., and triggers an alarm. The app generates SOS alerts, identifies user location and sends them to emergency contacts specified by the user.
This Rakshak Android app is the brainchild of Piyush Agrawal, Subham Banga, Aniket Sharma and Ujjwal Upadhyay from Bharti Vidyapeeth College of Engineering, Delhi. They started with publicly available speech command datasets, such as the Google Speech command dataset, then added speech commands specific to the scenario of women’s safety. They crowd-sourced additional data and open-sourced it as the Indian EmoSpeech Command dataset. This enabled them to detect emotion, background noise, and Indian accents in the audio with improved precision. The app is now available on the Google Play store.
The solution took the top prizeof $1500 last week in a contest organised in India by the Marconi Society’s Celestini Program. The project, run by the Society’s Young Scholars, is a flagship effort to inspire and connect individuals building tomorrow’s technologies in service of a digitally inclusive world. The Marconi Society and its Young Scholars select universities with promising telecommunications and engineering undergraduates and provide them with support and mentorship to help tap their true potential.
“This is the third successful year of the Celestini Program in India,” said Vint Cerf, Chair of the Marconi Society. “We see a clear trend of Celestini Program participants choosing research careers and technology-oriented graduate programmes, which helps us fulfill our mission of inspiring the bright minds that will bring the benefits of connectivity to the next billion.”
The team winning the second prize— also from Bharti Vidyapeeth College of Engineering, Delhi—addressed air quality issues since India has 14 of the 15 most polluted cities in the world, according to the World Health Organisation’s Global Ambient Air Quality Database. Team members Harshita Diddee, Shivam Grover, Shivani Jindal and Divyanshu Sharma created a privacy-aware smartphone app called VisionAir which uses photos of the horizon taken from a smartphone to estimate air quality. This builds on the work done by last year’s Celestini Prize winners which showed that a machine learning model can be built to estimate air quality from an image by extracting image features such as transmission index or haziness and combining them with meteorological data and historical air quality data. The innovative aspect of this year’s app is to leverage federated learning to train the machine learning model in a privacy-aware manner instead of uploading photos from each user.
Federated learning only uploads the features extracted from the images without uploading the smartphone images to train the machine learning model.
“Students become deeply engaged when they are defining the important problems that technology can solve and creating proof-of-concept applications that will make a difference in the world,” said Brejesh Lall, Head of Bharti School of Telecom Technology and Management and Celestini Program partner at IIT Delhi. The Celestini Program India partners with IIT Delhi and is anchored by Aakanksha Chowdhery, a researcher in Google Brain, and a 2012 Marconi Young Scholar.