Benefits of prior speech segmentation for best time-frequency visualisation using Renyi´s entropy

In this paper, a new approach that operates in the joint time-frequency domain for speech segmentation is presented. Segmentation is an important application in speech and audio processing. The segmentation in time domain is based on Renyi entropy especially on Renyi marginal entropy (RME) properties. Experiments were conducted using real-life speech signal as consonant-vowel (CV) transition that consists of two different events. They demonstrated the ability of the method for segmentation of speech signal made of CV transition. This technique is also useful for best time-frequency visualization with appropriate parameters. Because of the simplicity and effectiveness of proposed segmentation technique, it can be applied in many applications such as speaker identification/verification, estimation of the duration of the plosives, feature extraction, and classification.