Speaker Independent Continuous-Amharic Speech Recognizer: An Experiment Using Hybrid (HMM-ANN) System
No Thumbnail Available
Date
2004-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Automatic speech recognition is a part of natural language processing which
deals with making computers hear human speech and to take action according to what
they heard. In line with this concept, Speaker Independent Continuous Amharic Speech
Recognition is investigated to check the results that can be reached by an ANNIHMM
hybrid approach.
To implement the Amharic ASR system , 100 speakers, each speaking ten different
sentences have been modeled with the help of the CSLU Toolkit . The model is constructed
at a context dependent phoneme sub-word I eve!. A promising result of 78.56% word
accuracy and 44.07% sentence recognition rate has been achieved for speaker dependant
test and 74. 28% word accuracy and 43.70% sentence recognition rate for speaker
independent test. These are the best figures reported so far for speech recognition for
Amharic.
Description
Keywords
Information Science