The Coming Wave of Gadgets That Listen and Obey

dgan brings us a NYTimes piece about the development of speech recognition for common gadgets. Companies such as Vlingo and Yap are marketing their software to cellular carriers to give consumers a hands-free option for tasks like finding directions and text messaging. Quoting: "Vlingo's service lets people talk naturally, rather than making them use a limited number of set phrases. Dave Grannan, the company's chief executive, demonstrated the Vlingo Find application by asking his phone for a song by Mississippi John Hurt (try typing that with your thumbs), for the location of a local bakery and for a Web search for a consumer product. It was all fast and efficient. Vlingo is designed to adapt to the voice of its primary user, but I was also able to use Mr. Grannan's phone to find an address. The Find application is in the beta test phase at AT&T and Sprint. Consumers who use certain cellphones from those companies can download the application from vlingo.com."

Cell phone vs. server farm

[Lachryma]

The recognition that you describe is poor because the speech recognizer is running on the phone in a tiny memory/cpu footprint.

Most of the cell phone systems described in the article are likely uploading the audio to a server farm, running recognition there, and then sending back the response.