Sample-Based Automatic Dictionary Generation for Keyword Spotting System

In this paper we develop an approach to automatic, data-driven generation of pronunciation dictionaries for keyword spotting(KWS) systems. In practical applications, KWS tasks often have to deal with keywords whose pronunciations can not be found in the dictionary. To solve this problem, we study how to derive pronunciations automatically from speech samples of keywords. Recognized sequences from these samples are used as candidates, and merged to form a phoneme confusion network(PCN) from which the pronunciations are extracted based on a confidence-based metric. Experimental results show that sample-based dictionary reaches similar performance with the canonical dictionary, and the proposed approach is independent of the sample set.