My name is Daniel May, and I have been a member of ISIP since 2001. I recently finished my Master's thesis and have developed, as part of that work, a spoken term detection tool. Spoken Term Detection (STD) is the problem of searching vast amounts of recorded speech data for specific keywords. One technique for indexing audio is phonetic indexing. This technique does not suffer from out-of-vocabulary (OOV) word issues since the words themselves are not indexed. Instead, the phonetic information within the audio is indexed, and OOV can be handled in the search phase.

Our STD tool, which is part of the latest Prototype System release, uses phone lattices generated by our speech decoder tool as a search space for lexical definitions for a given set of words. All unique occurrences of the word are reported along with their start and stop times and normalized likelihood score. A demo experimental setup can be downloaded here. Detailed instructions for running this demonstration can be found in the AAREADME.text file at the top level of experimental setup.