Multimedia Keyword Spotting (MKWS) Using Training And Template Based Techniques

Abstract—The amount of multimedia information available is increasing due to the development of multimedia applications. As a consequence, content based searching and retrieval tools are needed for easy access to multimedia files. This paper presents a keyword spotter for searching a spoken word in a multimedia file using two techniques, Hidden Markov Model (HMM) and Dynamic Time Warping (DTW). To be useful, such a keyword spotting system has to be speaker-independent. Moreover, it has to be able to detect a word from a large vocabulary. This directly implies the use of a phonemic representation of the word which is achieved through HMM. Although in some scenarios, HMM approach is time and resource exhausting and hence DTW is used. MKWS can be applied to media files which can help in their automatic analysis and traversal