Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

The Application of Dynamic Programming Techniques to Non-Word Based Topic Spotting

P. Nowell, Roger K. Moore

DRA Malvem, Malvern, Worcs, England

This paper describes the application of dynamic programming (DP) techniques to the problems of building and testing non-word based topic spotters. We use a DP algorithm to find sets of similar phoneme sequence fragments, which we call DP-n-grams, and to detect their occurrences in the training and test data. The ability to use partial matches means that that the fragments are longer and more meaningful than the phoneme n-grams that have been tried previously. Detection probabilities of over 90% with less than 10% probability of a false alarm are achieved for seven target categories. Reports about bridges and pontoons are detected with 90% probability at a probability of false alarm of less than 1%.

Full Paper

Bibliographic reference.  Nowell, P. / Moore, Roger K. (1995): "The application of dynamic programming techniques to non-word based topic spotting", In EUROSPEECH-1995, 1355-1358.