Voice conversion using linear prediction coefficients and artificial neural network

Abstract

This paper describes the implementation of voice conversion system using linear prediction coefficients (LPC) and artificial neural networks (ANN). In this work, microphone recorded parallel speech utterances are used for voice conversion. Linear prediction coefficients are extracted from the speech utterance of the source and target speakers using linear prediction analysis. Auto associative neural networks (AANN) are configured and used for the transformation of LP coefficients from the source speaker to target speaker. LP residual obtained from target speaker and transformed LP coefficients are used to reconstruct the transformed speech. Mean opinion score and student t-test tests are conducted to evaluate the performance of voice conversion system. Evaluation of voice conversion system is performed in all 4 cases; such as male-to-male, male-to-female, female-to male and female-to-female. The test results show that the quality of transformed speech is good and AANN's have properly transformed LP coefficients of the source to that target speaker.