August 12, 2005

The Jerk-o-Meter

Today's news includes this article about the "Jerk-o-Meter", described as a device which can tell you whether the person you're talking to on the phone, or you yourself, is paying attention. It analyzes speech for features that reflect "activity" and "stress". The same technique provides better-than-chance predictions of the outcome of speed dates.

How useful a device this is is not so clear. Anmol Madan suggests that it might help to improve relationships by preventing arguments. Maybe, but I'm not so sure. Is receiving confirmation that the person you're talking to is not interested really going to improve the relationship? And note that it doesn't predict the outcome of speed dates in any useful way: its "prediction" is based on the conversation during the date, so it doesn't save you any time or angst.. Its kind of like the situation in weather prediction about twenty years ago, where they could predict the weather three days in advance but the computer model took three days to run, only in this case the problem won't disappear with faster computation.

Another suggested use is that:

it might assist telephone sales and marketing efforts
I thought technology was supposed to improve our lives. As an antidote, I recommend the National Do Not Call Registry.

If you read the paper that underlies the press reports, it turns out that there is something interesting here.

we propose that minute-long averages of audio features often used to measure affect (e.g. variation in pitch, intensity, etc.) taken together with conversational interaction features (turn-taking, interrupting, making sounds that indicate agreement like 'uh-huh') are more closely related to social signaling theory rather than to an individual's affect.
In other words, these features aren't uncontrollable subconscious cues to the speaker's mental state but controllable aspects of communication.

Posted by Bill Poser at August 12, 2005 10:12 AM