IDENTIFICATION OF FREQUENCY DOMAIN USING QUANTUM BASED OPTIMIZATION NEURAL NETWORKS

Dhivya bharathi, T.Karthikeyan, P.Hemalatha, P. Joy kinshy

Abstract: Voice Conversion (VC) is a zone of speech processing that contract with the conversion of the apparent speaker identity. This paper presents a new conversion method for text to speech and Voice to voice conversions. Text to Speech conversion uses phonematic concatenation for conversion and Voice conversion uses MLP Quantum Neural Network for transformations. The proposed method consists of two stages. In first stage, Features extraction of LSF and pitch residual based data from source and target speaker. In second stage where we use MLP Quantum neural network which is trained to learn the nonlinear mapping function for source to target speech transformation using the features extracted in the first phase. The proposed method can efficiently increase the quality and lack of naturalness of the converted speech. The proposed model is tested by male and female speakers with average duration.

Keywords: Voice Conversion, Line Spectral Pairs, Quantum Neural Networks, Multi-layer Perceptron’s, Phonematic Concatenation, TTS, and GMM.

DOI: https://doi.org/10.15623/ijret.2014.0319079