ESSA (Enhanced speech synthesis approach) for Building Punjabi Voice Model

Abstract

This Paper presents the text to speech synthesis model using Random Forest Technique along with mixed excitation approach for decision making. Base model is developed by extracting the various voice features types (segment features, phoneme identity etc.) in statistical parametric synthesis approach, which is further enhanced with Random Forest criteria to redevelop the voice model. Twenty cluster trees are generated in Random forest from which one best is selected and used to create a voice model.In this paper for each developed text to speech model, the Mel-cepstral distortion scores are evaluated for comparative study.