Siteseeker Voice - A speech controlled search engine for Swedish

SiteSeeker Voice -
A speech controlled search engine

Hercules Dalianis, Adam Blomberg,
Rolf Lindgren*, Johan Carlberger±
Martin Hassel

NADA-KTH
Royal Institute of Technology
100 44 Stockholm
ph: +46 8 790 91 05
mobile: +46 70 568 13 59
email: hercules@kth.se
* Presector AB
± Euroling AB

In the Wapalizer project supported by Vinnova, we have constructed one of the first speech search engines called SiteSeeker Voice. One can search for information on a web site using a speech interface. This interface is accessed by phone using a special telephone number. A voice will direct you on how to search for information on a specific web site. The user can say search words and command words to find the information. The found information is extracted/summarized and read by the synthetic voice.
There is a similar approach carried out by Google called Voice Search
http://labs1.google.com/gvs.html but to see or read the results you have to use a web-browser. In the Google approach you are tied to a computer while with SiteSeeker Voice you are completely free.

SiteSeeker Voice is based on Eurolings search engine SiteSeeker, http://www.euroling.se, and the voice platform speechWeb from Pipebeach, provided by the company Voxway. The application was designed in cooperation with Presector Speech Technology.

speechWeb make use of VoiceXML. We have extended SiteSeeker that it generates VoiceXML that is interpreted by the speechWeb application. The license for speechWeb delimits to a maximum of 500 recognized words for each search. We have selected these 500 words from the most common search words from the statistics from each customer, but also as the words that is the most relevant following information retrieval theory, i.e. word with the highest IDF.

SiteSeeker is used by over 20 muncipalities and public services in Sweden. The same search engine SiteSeeker is used by each site is also used by SiteSeeker Voice. The difference is that one can not yet select other input language than Swedish and that you only get the first four hits read on each search. Spell checking, selecting of advanced search is also disabled.

SiteSeeker Voice is only a prototype.

You can call the telephone number +46 (0)8-598 96 731 and talk with the search engine SiteSeeker Voice, say first the web site you want to search on, e.g. Skurup or Vinnova and then you search one of these sites.


Here you can see all the available sites, though not all tested.
http://default.siteseeker.se/voice/

Here you can see all the allowed words of the VoiceXML search of Skurups kommun (muncipality) http://skurup.siteseeker.se/voice/

Here you see the answer of the word "bibliotek" applied to SiteSeeker Voice for Skurups kommun
http://skurup.siteseeker.se/voice/?query=bibliotek

Applications

Improvements that are needed


Responsible for this page: Hercules Dalianis <hercules@nada.kth.se>
Latest change February 25, 2003
Technical support: <webmaster@nada.kth.se>