Vlingo CEO Dave Grannan told me the thing that's holding back the growth of mobile applications is the hassle of using that little keypad on your device. Agreed. Grannan's company announced its answer to that problem yesterday, a new mobile voice recognition software that enables mobile devices to understand your verbal commands.
So here's how it works. Vlingo isn't just a piece of software that lives inside cell phone, its brain is located at a bank of servers in Massachusetts. When a mobile user speaks a request into the phone, the sound is sent to Vlingo's servers where it is recognized and turned into text, then sent back to the mobile device.
I tried using Vlingo to search for music at Sprint's Music Store, then did some voice-triggered local search using Vlingo Find. For both services, when the search box comes up, you hold down the Talk button on your phone then begin speaking your search terms.
When you let up on the Talk button, Vlingo displays what it heard you say. If the servers get it wrong--if you say Sanjaya and Vlingo hears Sangria, for instance--the user then corrects the erroneous letters or words using the keys on the phone.
Grannan says Vlingo then notes its error, and learns to make the correct association between the sound "Sanjaya" and the word "Sanjaya" in the future. If some other user says "Sanjaya" to Vlingo, it will get it right. So, in theory, the more mobile users talk to Vlingo, the better it gets at understanding.
At Sprint's Music Store, Vlingo easily recognized and found results for "My Chemical Romance," "The Rolling Stones," and "Nick Cave and the Bad Seeds," but when I said "David Hasselhoff" it heard "David Hassle of," then "David Hassle Hoff." I keyed in the correction twice, then said "David Hasselhoff" once again, but the system still hadn't "learned" the name. I wouldn't want to know David Hasselhoff's name either.
Vlingo Find is supposed to return local search results based on your verbal commands. To test this, I went outside where a good amount of street noise was present. Vlingo had no trouble understanding and returning relevant search results for commands like "McDonalds restaraunts in San Francisco". Once you pick out the search result you want, you can tell Vlongo to plot it on a local map for you.
Things went a little rougher when my search terms were less specific, however. When I spoke "San Francisco, First and Market, record stores" into the phone, Vlingo understood my command almost perfectly (only one letter was wrong), but the search results came up empty. Same thing happened when I told it "San Francisco, 501 Second Street, pizza." That, however, might be a problem with the local directory listings Vlingo Find uses, not the voice recognition system.
All in all, I was impressed with Vlingo's ability to recognize my words. When it got a couple of letters wrong, it wasn't too painful keying in the corrections. I'm still not convinced, however, that Vlingo wouldn't repeat those same errors.
Vlingo intends to sell its voice recognition service to wireless web applications developers and to wireless service providers. So far, Grannan says, his company has an agreement only with Sprint, but a deal with AT&T is in the works.
