A Method to Define a Search Query by Speech

Making the content of web pages more accessible for visually impaired people involves displaying web pages sonically mostly by speech. The other aspect of web browsing entering information into web forms remains nearly untouched. This aspect is one of the biggest advantages of the internet itself the possibility to use search engines Google, etc. Before searching the Internet, the user must enter a query, which is composed of a list of isolated keywords. To recognize a spoken query for a search engine, it is necessary to recognize a linear list of keywords separately. As these can be nearly any combination of words or abbreviations, the recognition is a difficult task for any current speech recognizer. In this paper, we describe an efficient method to enter a query by speech. Our method is based on predicting a list of keywords for the next word (which is the user going to dictate) from precedent keywords. The fact, that as a result of this prediction there is only a limited set of keywords to choose from, dramatically increases the precision of the recognizer. The algorithm itself is based on cluster analysis and is applicable for virtually any database query.