Voice Conversion by Advance Method

Voice conversion is nothing but the modification of the characteristics of voice signal of one speaker (source speaker) so that it sounds as if it had been pronounced by different speaker (target speaker). In this method of Voice conversion we use the spectral characteristics such as frequency, power spectral density, amplitude etc. As compare to other methods this method provides more efficient transformation because in this method we are using more than one spectral characteristics.

Since, here we are using many characteristics of the signal so we can say that it is an implementation of frequency-warping-based  voice transformation in which only the frequency of both source and target speaker are extracted and modify. Evaluation by objective tests and formal listening tests show that the proposed transform method greatly improves the quality, naturalness as well as noise information of the converted voice signal compared with other proposed transformation methods.

Over the last few years, the interest in voice conversion has risen significantly due to its wide application. Two main aspects of the transformation problem are voice quality and intonation.

In this paper, we focus on the control of noise which present in the voice. More significantly, our aim is to represent by an appropriate model, trained from experimental data, the statistical relations between the spectral envelopes of two different speakers uttering the same text. To differentiate this problem from the general voice transformation task, which would also necessitate a proper analysis and control of the characteristics of signal, we will refer to the control of the spectral envelop as spectral transformation.