Abstract
Automatic reference point placement method for voice morphing is reported in this paper. Voice morphing is one of fundamental voice editing methods to blend feature vector sequences of two voices based on corresponding reference points. Reference points are basically assigned by hands, and depends on the quality of voice morphing output. Moreover, assigning reference points is a time-consuming task. The proposed method realizes to assign reference points on spectrogram in time- and frequency-domain automatically based on temporal decomposition (TD) and line spectral frequency (LSF). As results of two-speakers’ voice morphing, the proposed method was worked well by using voice and its transcription as inputs.
