Speech Unit Selection Based on Matching Pursuit

This paper introduces a new method based on Matching Pursuit for speech unit selection. We used the Matching Pursuit transform parameters with a comparison algorithm to find the best match for a selected unit in a Text-To-Speech system based on concatenation. We chose Gabor atoms. Also Wigner-Ville distribution implemented for the time-frequency presentation of the transform and we used image processing approach to compare these time-frequency presentations of the acoustic units. On a database of 42 units 92% accuracy was obtained.