Abstract:
Speech recognition is the process of converting an acoustic waveform into the text similar
to the information being conveyed by the speaker. This paper aims to discuss the design and
development of isolated word Afan Oromo speech recognition system which can be used to
command and control Fugugaa Afan Oromo word processing software. In order to develop the
desired system development of speech corpus was one of the fundamental requirements. Thus, the
research has initially collected 59 distinct Afan Oromo command from Fugugaa Afan Oromo word
processor based purposive sampling method. Having the command words Afan Oromo speech
corpus was developed. Speech data of those words were recorded from 20 individuals of which
the number of male were 12 and the remaining 8 were female. The speakers were asked to utter
the words 3 times. So the 59 command words resulted in (59*3) samples of 20 distinct
speakers files making a total of 3540 (59*3*20) files. The speech data were divided into two
parts: training and test set. The training set which was used to train the recognizer has used 2478
files while the test set which was used to test the performance of the recognizer has used the
remaining 1062 files. The research conducted develops a small vocabulary speaker independent
isolated word recognizer for Afan Oromo language. The HTK toolkit based on Hidden Markov
Model (HMM), a statistical approach, was used to develop the system. The performance of these
recognizers was tested using both the test data and the live recognition. Although the performance
of the recognizer using the test data performs very well for which it recognized all the test data
correctly, the performance of recognizer with live recognition tested using eight different speakers
for four of them were involved in the training of the recognizer while the remaining were those
who were not involved in the training of the recognizer performed average of 92.8%.
In general, according to the performance analysis of the recognizer, it indicates that it possible to
develop HMM based isolated word Afan Oromo speech recognition system which could be used
as speech input interface in order to command and control Fugugaa Afan Oromo word processor.
Based on the result of the study and what has been learned in the course of the research
recommendations for further study were forwarded.
