A Continuous, Speaker Independent Speech Recognizer for Afaan Oromoo
No Thumbnail Available
Date
2010-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
The ultimate goal of any automatic speech recognition is towards developing a
model that converts speech utterance to texts words. Therefore, a continuous,
speaker independent Afaan Oromoo speech recogniser’s experiment is performed
having similar objective of transforming Afaan Oromoo continuous speech in to its
text word formats for continuous Afaan Oromoo speaker independent speech
utterances using HMM and sphinx system (sphinx train for training and Sphinx4 for
decoding). Therefore, this research tries to develop prototype for a continuous,
speaker independent Afaan Oromoo speech recognizer so as to check possibility and
suitability of the tools and techniques selected from the various literatures.
A continuous, speaker independent Afaan Oromoo speech recognizer is developed
in this research work for 70 selected Afaan Oromoo long words, phrase and simple
sentence uttered by 30 selected peoples from different age group and sex
constituting of 2100 utterances. Accordingly, the data collected was divided in the
2/3 by 1/3 for training and testing respectively. Furthermore, various preprocessing
and other activities were performed including building the acoustic and
language models among others which might greatly affect significantly the
performance of the recognizer. These Afaan Oromoo selected words, phrases and
simple sentences are selected in consultation of the domain experts.
For this research performance evaluation is performed using test data sets and the
recognizer performance is found to be 68.514% with sentence accuracy of 28% for
continuous Afaan Oromoo speech and a phoneme based trigram performance of
89.459% with sentence accuracy of 42% achieved.
According to the finding of this research, the performance gained for Afaan Oromoo
language is highly promising and as the language is becoming one of the most
spoken language of the country, it will have tantamount for latter full deployment of
the recognizer in the language.
Keywords: Speech recognition, Afaan Oromoo, sphinx, Hidden Markov Model
Description
Keywords
Speech recognition, Afaan Oromoo, sphinx, Hidden Markov Model