About us




Contact us:

Email: office@hnny.biz

phone: +972 8 9472771

fax: +972 8 9470577


Engine for search in large audiofiles.

Locating specific information in large databases is a challenging problem. Particularly in case of searching for fragments of files possessing some specific properties. If the file is a text file, one of the options is search by key words and expressions. This technique is very well developed.
A much more complicated problem appears if the given file is not a text file but an audio file which contains a record of natural human speech. The problem of search in large audiofiles, especially 'in real time', is very important for many applications: for business executive, media analysts, political commentators, security services. This problem is extremely difficult because of the high variability of human speech.

The 'straightforward' approach to this problem is first to transform speech to text using some speech recognition system, and then to perform the search in the obtained text file. The second part is not a challenge now: the search in text files is not difficult problem. However, the first part of such a program - the automatic speech to text transformation - can not be considered a solved problem. Speaker-independent speech recognition programs work very slowly, even for the best recognition systems the time of recognition is 85% longer than the total duration time of speech.

Moreover the recognition reliability of speech recognition programs is not satisfactory -not higher then 70%. This non-reliability may result in loss of searched words or expressions. For some applications, such a loss is not acceptable.
We've developed an innovative high speed search engine for search by key words in massive audio files. It does not use any speech recognition system, works incredibly fast (time of search in a voice file is not more then 5% of the total duration of the file. It is not dialect-dependent and has the capacity to accommodate any pronunciation


Copyright 1998-2012 hnny.biz