April 12, 2006 1:07 PM PDT
Google patent points to voice search
Patent No. 7027987, of which Google is the assignee, concerns "a voice interface for search engines. Through the use of a language model, phonetic dictionary and acoustic models, a server generates an n-best hypothesis list or word graph."
Though the patent was published Tuesday, the application was originally filed in February 2001, indicating that Google has had the project in the works for some time.
A demo of something called Google Voice Search has been up on Google Labs, Google's pre-beta-test site, for well over a year.
Craig Silverstein, the director of technology at Google, said in a 2004 interview, that the company envisioned a voice interface to aid in everything from driving directions to finding groceries in a supermarket.
But on Wednesday, a Google spokesman cautioned against reading too much into the publication of the patent.
"We file patent applications on a variety of ideas that our employees may come up with," Barry Schnitt said. "Some of those ideas later mature into real products or services, some don't. Prospective product announcements should not be inferred from our patent applications."
Google Voice Search, still up on Google Labs, lets people call into Google by phone.
Though the demo on the Web page was inactive when CNET News.com tried to test it, the instructions show that this is how the system is meant to work:
Searchers call a Google supplied number, where a prompt asks them to "Say Your Search Words." After a minute, the results are returned and, in the demo, the searcher clicks a link that leads to them. In a real-world application, the results would presumably be sent to a cell phone screen or in-car system.
The obvious eventual application is for cell phones. Mobile Web services are on the rise, and cell phone keyboards could make typing in keywords a nuisance.
Google isn't the only one to do work in this area.
Another voice interface was released in May of 2005 by a team led by Dr. Meirav Taieb-Maimon of the Department of Information Systems at Ben-Gurion University of the Negev in Israel. Her team's voice-activated Internet system works with speech-to-text conversion software called Maestro.
Maestro converts spoken search requests into a text of query-friendly words that it sends to a search engine, and relays results back to the searcher audibly. In 2004, Silverstein called that latter capability an obstacle to successful voice interface technology.