Handling Pronunciation Variation Using Hybrid Approach in Continuous, Speaker Independent Speech Recognition for Amharic

No Thumbnail Available

Date

2014-10

Journal Title

Journal ISSN

Volume Title

Publisher

Addis Ababa Universiy

Abstract

The problem of modeling promllciation variation lies in accurately predicting the word pronunciations that occur in the test material .In order to achieve this, the pronunciation variants must first be obtained in some way or other i.e. from pronunciation data or from pre-specified phonological rules based on linguistic knowledge. In this study, first models was developed using canonical dictionary as a reference with total data set of 950 sentences of which 700 for training and the remaining 250 are used for testing model Two models were developed using knowledge based and data driven adding variants to the respective dictionaries. Lastly another model: hybrid approach was developed. This model is supposed to avoid the short coming of the two models, knowledge based and data driven. In light of this, the model developed using hybrid approach has shown better performance. The better progress obtained is due to the fact that in case of knowledge based approach, the variant added mayor may not really appear in the text audio. in contrary, during data driven the variants actually appear in audio but it is very difficult to listen and identify all words with variation. Thus the experiment undertaken in this study revealed that combining the two approaches (hybrid) is the best way for handling pronunciation variation.

Description

Keywords

Speaker Independent Speech Recognition for Amharic

Citation