17-09-2009, 02:14 AM
ABSTRACT
The evaluation of text-to-speech system can be started evaluation of speech generated by the system. There are two evaluation criteriaâ„¢s that are neutrality and intelligibility of speech. Theoretically, articulator synthesis is a good method of speech-generating process for the text-to-speech system because it can generate natural speech. So, it will have high intelligibility rate. Until now, there is no complete data of articulator-model parameters for generating Indonesian phonemes. Whereas, these data are very important in Indonesian text-to-speech system development which uses articulatory synthesis for generating speech Therefore, a research on finding these parameters has to be done.
In this research, parameterizations involve speech synthesis process which is done using ARTM toolbox in MATLAB. Speech synthesis process itself can be divided into analysis, speech inverse filtering, excitation set up, and synthesis process. Pre-processing and post-processing can be added to those processes. Pre-processing provide formant contour of speech signal from the speech recording input. Frames which contain four first formant frequencies information is produced from formant contour in analysis process. Articulatory-model speech generator parameters for certain frame then can be provided from the frame in speech inverse filtering process. Using these articulatory-model parameters and excitation parameters, which are determined in excitation set up, synthesis process produces speech signal. Post-processing is done to determine the right frame to represent a certain phoneme and its articulatory-model parameters.
Using this methodology, articulatory-model speech synthesizer parameters for almost all Indonesian phonemes, except fricative phonemes, can be obtained. Although those parameters have not represented ideal articulatory-model yet, articulatory-model speech synthesizer parameters for Indonesian phonemes which are obtained in this research can generate human-speech-like signal