Speech Signal Segmentation using Zero Crossing Rate and Short-Time Energy for Speech Synthesis

The selection of unvoiced and voiced signals is typically made through the analysis of speech by extracting the desired set of information from the given speech signal. The different approaches for distinguishing between the segments of unvoiced and voiced portions speech signal are presented in this study. These techniques rely on the autocorrelation of various speech signal segments as well as the zero crossing rat (ZCR) and short-time energy (STE) calculations. Theoretical research has shown that spoken segments have high energy and magnitude whereas voiced signals have low ZCR rate. In this work, the autocorrelation function is used to demonstrate how unvoiced signals lose their periodicity while the voiced portion of speech maintains its periodicity after applying the autocorrelation function. The simulation results depicted in this paper validates the theoretical research.