A group of University researchers has invented and produced MusicalHeart, a smartphone app that responds to the user’s heart rate, activity level and context to recommend music during a range of activities.
The app uses a set of sensor-integrated headphones that measure acceleration and heart rate to gauge the listener’s activity level and then report to a remote server that recommends music to the listener to maintain a target heart rate. The initial prototype was completed in April, and the second version containing better screening technology was finished this month.
Shahriar Nirjon, a graduate student in the department of computer science, said he had always wanted a device he could use to generate a soundtrack for his day-to-day activities. “Whether I am driving, jogging, traveling, or relaxing, I never find the appropriate music to listen to.” Nirjon said in an email. “The problem is the heart wants to hear something but our music player does not understand the need. My joy was in connecting them together in a non-invasive and cost-effective way.”
MusicalHeart aims to fill that need. It reads the heart rate, activity level and geographical context detected by the sensors in the headphones, combines the three and inputs the information into the online database, which then recommends music relevant to that specific activity.
To collect data for the project researchers tested 37 participants taking part in low-, medium- and high-energy activities so they could observe the correlation between music and heart rate. The researchers then came up with three separate algorithms: one for detecting the signals from the ear to translate to heart rate, one for detecting activity levels and one for the music recommendation system via biofeedback.
Gathering this data was not easy, Nirjon said.
“There were not many participants who were willing to run for an hour wearing all sorts of physiological sensors attached to their body,” he said.
The study, “MusicalHeart: A Hearty Way of Listening to Music,” said the accuracy of the person-specific activity level detector is close to 100 percent, on average.
Though the app has yet to be released in any app store, the project study was published in the 10th ACM Conference on Embedded Networked Sensing Systems. “We will also be doing a public demonstration of an improved version of our hardware at the demo session of the conference,” Nirjon said. “Currently, we are improving our system and exploring new possibilities with our hardware prototype.”