Speech synthesis using artificial neural networks

Abstract

Statistical parametric synthesis becoming more popular in recent years due to its adaptability and size of the synthesis. Mel cepstral coefficients, fundamental frequency (f0) and duration are the main components for synthesizing speech in statistical parametric synthesis. The current study mainly concentrates on mel cesptral coefficients. Durations and f0 are taken from the original data. In this paper, we are attempting on two fold problem. First problem is how to predict mel cepstral coefficient from text using artificial neural networks. The second problem is predicting formants from the text.