Conversion of one emotional state to other of a speech signal using Artificial Neural Network

Abstract

This paper presents a novel emotion transformation scheme of speech signal which is text independent and speaker independent. Speech signals as many other signals are inherently multi-scale in nature, owing to contributions from events occurring with different localizations in time and frequency. Therefore, emotion dependent spectral parameters those characterized by single scale features, approximate the vocal tract, but produce artefacts during speech signal reconstruction. In this paper, multi-resolution spectral transformation technique of Discrete Wavelet Packet Decomposition has been used along with the use of Artificial Neural Network for generation of transform function. This paper specifically carries out transformation of Neutral emotion to Angry, Happy and Sad emotions. The transform function is generated in three different techniques, using three types of Artificial Neural Networks (ANNs), namely, Feed Forward Neural Network (FFNN), Generalized Regression Neural Network (GRNN) and Radial Basis Network (RBN). Results of all the three ANNs are compared using both objective as well as subjective analysis.