Novel Noun Pronunciation Unification Approach to Improve Story Boundary Identification in the Transcription of Malay News Broadcasts.
Zainab Abbas Khalaf, Tan Tien Ping · International journal of computer science and applications · 2014
This paper introduces a novel method for improving story boundary identification (SBI) for a speech recognition outcome. We explore an SBI improvement method using latent semantic analysis (LSA) with noun unification. The proposed system uses the phonetic forms of words (pronunciation forms) to identify story boundaries based on noun unification and edit distance to estimate the cost of edit operations for nouns and to compare this cost with a predetermined threshold generated by a training dataset. SBI commonly uses latent semantic analysis for its excellent performance and because it is based on deep semantics rather than shallow principles. In this study, the LSA algorithm with and without unification was used to identify the boundaries of Malay spoken broadcast news stories. The LSA algorithm with the noun unification approach resulted in less errors and better performance than the LSA algorithm without noun unification. The preliminary results of the current work for SBI using LSA with noun unification are encouraging compared with the common LSA with the general approach for selecting bag-of-words.