Tuesday, November 1, 2005


Ajay Juneja's company, Speak With Me, is nicely profiled in TechCrunch. Since I worked at a speech technology company, I wasn't as impressed as Michael from TechCrunch or Robert Scoble was when I experienced the demo a couple months ago. The command and response activities are pretty straight-forward. Some of the more advanced functionality is based on educating the system. At HeyAnita, for demo purposes, we set it up where we could say to our voice-activated information service, "I want to fly from Los Angeles to Chicago on May 11, 2000 and come back on May 14, 2000." Lots of information and lots of commands, which impressed people, but in the real world this would probably have a verification rate of 50%-60%. The demo and our team learned how to interact with each other, so it was almost 100%. For voice recognition services, you want at least 95%.

What Ajay's technology does well is the ability to identify syntax and to provide multiple commands to his system. The system has learned how to recognize the hundreds of songs he put into his system. His demo for searching and playing various songs in his system is what I was impressed by. Also I was impressed that he has taken it into the chipset, so eventually he can expand his product beyond cars to mobile devices and phones. Anyway, it is a great product and application he's built.

No comments: