Search This Blog

Monday, May 9, 2011

IBM has created a search engine for the illiterate

Previously, IBM has developed a network of voice sites for residents in several Indian states, as well as Thailand and Brazil. But now this curious project has entered a new stage: experts have come up with a voice search engine to navigate the sprawling network

Now the unusual similarity of the Internet called Spoken Web used by about 10,000 people, but in future network-driven voice and gives information only in the form of sounds, can cover a fifth of the world's population. Experts estimate that just as many people in the world can not read.
The inhabitants of the countries listed above use the numbers as an analogue of addresses on the web. Dialing a certain number, they get to voice website that can learn, for example, recent prices for grain, or listen to the announcement of the available vacancies.
"As the number of sites and increase the voice of their content there is a problem: people need to quickly navigate and find what they need," - explains Nitendra Rajput (Nitendra Rajput), fellow IBM Research India, one of the founders of Spoken Web.
Now anyone who wants to create a voice site, it is proposed to come up with his name and share it on the information sections. To navigate in them, a person uses an automated telephone system that accepts voice commands.
But to listen to a lot of unnecessary information, tedious and, moreover, is expensive, not to mention the fact that you must first select the site of a dozen (sometimes more) alike.
Voice Internet works in trial mode for four years. In the first 8 months of a pilot project in rural India, the services of the oral network "used by more than 6,500 people visited the relevant sites are about 114 thousand times (Figure Neeraj Yuvraj / Flickr.com).



To expedite the process, developers have created a new engine. The classic version of voice search will not work. The man who spoke in the name of the phone you want him to pesticide will not listen to transfer 20 search results. He simply did not remember them.
So, we need to narrow the possible request parameters and to reduce the output information for five or fewer points, the authors decided the system. Now the user is asked to filter out the right: to name the person who created the Voice website, the place where it was created, as well as to clarify the section, which was placed the site (for example, choose between "news" and "questions and answers"). Of the five results to select the desired portal much easier.
Now new technology is being tested at farmers in the Indian state of Gujarat. IBM employees are ready to offer a new tool for all users Spoken Web, but note that further development of networks requires new solutions.
For example, to accelerate the search process can be "smart" system, which dissipate the sound information in an accelerated 10-fold mode, stopping only on the keywords request. Now people can take that rewind, but stopping it is necessary to do yourself.
Would be good to train the system itself to determine which words or phrases are important for the listener. To do this you need to dial the statistics and analyze users' behavior, stop and accelerate the "read" websites.
Now Spoken Web, in fact, has no relation to the World Wide Web. In addition, users voice network, usually for local websites and news. But the developers at IBM are hoping that in the future (at least to improve search engine), the information from the "general" of the network still will appear in your voice.
"We can transfer data using the API-calls, and technologies that convert text to speech - says Rajput. - But they will also need to be translated into the language, and with it outside of the English part of the Internet problem. "

No comments:

Post a Comment