Spoken term detection from noisy input

Abstract

The aim of the spoken term detection task is to find the occurrence of user-entered keywords in an archive of audio recordings. The kind of techniques that are used usually are vocabulary-independent, using only the acoustic information available. In this scenario, however, we rely exclusively on the acoustic model, which is a drawback when it is unreliable; for example when the input is noisy. In this paper we investigate the possible accuracy of spoken term detection when the recordings were obtained using small-capacity, portable devices (wireless sensors) that have a quite low-quality microphone. The accuracy scores show, however, that despite the high amount of noise in the input recordings, our spoken term detection method can still produce an acceptable level of accuracy.