Searching a website via spoken questions

Abstract

To reach the desired information on a web site can be a very time consuming task when the fact that the site may not even have that information is considered. In addition to the time consuming feature of checking all the pages, the user can also overlook what she looks for which can be regarded as a human factor. For all these reasons, some automatic search engines are developed, at di erent success levels. In this thesis, a system is proposed which takes the spoken queries from users and forwards them to the page that has the best answer in order to save them to check all the pages manually. In addition, this kind of solution gives users the chance to communicate with the system by asking spoken questions. This is a very important advantage especially for the visually handicapped people. Another service of the system is to prepare the page to be read by the Text-To-Speech unit. In this way, people can talk to the system and nd answers to their questions in various platforms like Kiosk applications without any written command. After automatic recognition; the query is classi ed and keyword assignment is done. Then this keyword is searched in the target pages and nally the most suitable page's address is returned to the user. For the classi cation process, su cient amount of data is collected and the system is trained. After the classi er is constructed, the input is classi ed into one of the previously speci ed 110 classes. To sum up the entire system, the user interacts verbally with the system and is forwarded to the page which contains information evaluated as a candidate answer to the question. In short, users are supposed to ask questions requesting information, an answer, or a transaction the system is supposed to bring a relevant page accordingly.