Multimodal Voice Conversion Using a Deep Convolutional Neural Network in an Adverse Environment

This demo page provides some samples of speech conversion results using three different speech conversion algorithms in different noise environments. The proposed MDCNN, the traditional NMF and the DNN algorithms are used respectively to conduct male to female speech conversion (Demo 1) as well as female to female speech conversion (Demo 2). Six types of noise from the noise97 database are used to corrupt the clean speech to obtain the noisy speech in different noise environments.

Recommended browser: Google Chrome

Cite as: Jian Zhou, Yuting Hu, Hailun Lian, Huabin Wang, Liang Tao, and Hon Keung Kwan, "Multimodal voice conversion under adverse environment using a deep convolutional neural network," IEEE Access, vol. 7, pp. 170878-170887, 26 November 2019, DOI: 10.1109/ACCESS.2019.2955982