Transformation of sound by statistical techniques is a promising method for a new range of digital audio effects. In this paper a data driven voice transformation algorithm is used to alter the timbre of a neutral (non-emotional) voice in order to reproduce a particular emotional vocal timbre. Perceptually based Mel-Cepstral analysis and Mel Log Spectral Approximation digital filter are used to represent the speech timbre and to synthesize speech with modified spectral envelope. The transformation function adopts a GMM (Gaussian Mixture Model) based parametrization in order convert the spectral envelopes. Experiments with the first and second order derivatives of the mel-cepstral coefficients have been undertaken to prove the benefit of including dynamic information in the model. The proposed algorithm has been evaluated by means of objective measures in the neutral-to-happy and neutral-to-sad tasks.

Download