Hubert Ségot and Hisao Kuwahara
Voice Conversion by Analysis-Synthesis
Method
Abstract:This paper deals with a way of controlling the quality of natural speech by
manipulating such acoustic parameters as formant frequencies/bandwidths
and pitch frequency. These parameters were modified making use of the
conventional analysis-synthesis system. After extracting trajectories of the
lowest three formants, their frequencies and bandwidths at each analysis
frame were subjected to change. Modification of pitch frequency was performed
by changing the first peak of the spectral envelope obtained from re-analysis of
the residual signals.
This work was done as an internship at ATR Interpreting Telephony Research
Laboratories. Software programs for the above methods were developed during
the internship period. A few examples of synthesized speech by changing the
acoustic parameters are included.