Indexing and retrieval of speech documents

Abstract

In this paper, a speech document indexing system and similarity-based document retrieval method has been proposed. K-d tree is used as the index structure and codebooks derived from speech documents present in the database, are used during retrieval of desired document. Each document is represented as a sequence of codebook indices. The longest common subsequence based approach is proposed for retrieving the documents. Proposed retrieval method is evaluated using a speech database of 3 hours recorded by a male speaker and speech queries from 5 male and 5 female speakers. The accuracy of retrieval is found to be about 88% for the queries given by male speakers. © Springer International Publishing Switzerland 2014.