Wednesday, August 08, 2007

Making Phones Polite

Privacy-sensitive software could minimize interruptions.

By Kate Greene

Chatterboxes: This vertical line represents all the face-to-face conversations that took place among between the participants in the Intel study during the course of a day. The length of the wedge jutting from the line represents the length of the conversation. Here, all conversations except those of participant seven are grayed out. The other colors signify the size of the group chatting. An orange wedge represents five or six people, and a red wedge represents two or three. Credit: Beverly Harrison, Intel

Multimedia •  View images of the device, how it can be worn, and the data it produces.

There's no doubt that mobile phones are becoming more like computers as they gain functions, storage capacity, and processor speed. But even with Internet access and touch screens, these gadgets are still dumb machines. They know nothing about the person who uses them, and in particular, they don't know when it's appropriate to interrupt.

Now, researchers at Intel have developed software that could help make handhelds more considerate. The software is able to detect and record conversations, but crucially, it does so in a privacy-sensitive manner so that the actual spoken words can't be retrieved. "Our goal is to be able to collect data about interactions and conversations that happen spontaneously ... and have a balance between privacy and the information we can get from recorded data," says Tanzeem Choudhury, a researcher at Intel Labs Seattle.

The researchers capture information about how human speech is produced, says Choudhury, rather than information about the words themselves. Participants in the study wore a pager-size device with a microphone, some memory, and a processor that processed the audio so that features such as volume, pitch, tone, and rate of speech could be estimated. Surprisingly, Choudhury says, this information can speak volumes about a person's situation, mood, and social network. "You cannot get what is being said," she says, but "they allow us to make different types of inferences."

For example, the amount of time a person speaks and the number of interruptions in a conversation can indicate the status of the people in the conversation, Choudhury says. For instance, a boss-and-employee interaction would likely be a conversation in which one person does most of the talking and there are few interruptions. Conversely, if a conversation is dynamic, and there is a lot of overlapping speech, it's most likely a casual or social conversation. In addition, Choudhury says, people's mood can be inferred because speaking rate, loudness, and pitch change when they're angry, happy, excited, sad, and so on.

The Intel team isn't the first that has collected conversations to gain insight into social interaction, says Nelson Morgan, director of the International Computer Science Institute and a professor of electrical engineering and computer science at the University of California, Berkeley. For about a decade, researchers have recorded business meetings, he says, with the hope of analyzing the social structure and determining who made decisions, who was dominant, and what the key ideas of the gathering were. But not as much work has focused on trying to analyze impromptu conversations, and it is novel, Morgan notes, to consider privacy in the data-collection process. "That's a neat idea, and they did very well with it," he says.