Sometimes it鈥檚 better to share. of Dartmouth College in Hanover, New Hampshire, and his colleagues have created a system for pooling the plethora of data gathered from phone sensors such as microphones. Called Darwin, it could be used to improve the accuracy of the sensor data collected, the team says.
Phones that have the Darwin software installed on them learn to recognise an event, such as a specific person鈥檚 voice, and how that may change in different conditions, such as when that person is in a noisy room. The phones then share this information with other phones that contain the software.
To show that this sharing approach increases the quality of data, the team installed Darwin on the phones of eight people and tested a voice-recognition tool while they visited a quiet meeting room, a noisy restaurant and a busy street.
Advertisement
The handsets then attempted to identify a speaker using the voice recognition software available on the cellphone. The phones with Darwin installed on them compared their identification efforts in an attempt to improve accuracy.
Shareware
This sharing allowed phones with Darwin to correctly identify which person was speaking with up to 90 per cent accuracy, compared with just 60 per cent in a control group that didn鈥檛 have the software on their phones.
鈥淏y relying on the resources around you, you get better results,鈥� says Miluzzo. 鈥淭he idea is to make the sensing process on phones scalable.鈥�
While this application may have shades of Big Brother, Miluzzo says it could be used to improve other cellphone applications that rely on sensor data, such as GPS.
鈥淭his is a nice technique,鈥� says at the Georgia Institute of Technology in Atlanta. While similar data-sharing tools have been built for static sensor networks, they have been hard to perfect for mobile devices as the processing power needed can impact battery life.
The work was presented at the conference in San Francisco last month.
