Fast Unconstrained Audio Search in Numerous Human Languages

Abstract

We present a system to index and search conversational speech using a scoring heuristic on the expected posterior counts of phone n-grams in recognition lattices. We report signi cant improvements in retrieval effectiveness on ve human languages over a strong 1-best baseline. The method is shown to improve the utility (mean average precision) of the retrieved lattices' rank order and to do so with a search cost negligible compared to the fastest yet known methods for the linear scanning of phonetic lattices.

Fast audio search using vector space modelling

“…As in previously reported systems for audio search [1,2,3,9,10] we use a lattice representation for each spoken document in our database. For vector-space modelling, it is necessary to extract an unordered list of terms of interest, along with their counts, from each document in the dataset.…”

“…In this paper we use LSA to sort a list of all of the audio segments in the database, according to the criterion in (9) for each of the 1107 queries provided by the NIST Spoken Term Detection evaluation initiative. The sorted lists are then truncated and searched in the second stage of our spoken term detection system.…”