A Study on Voice Conversion Using Speech Synthesis Based on Short-Time Fourier Transform Magnitude

摘要

This paper proposes a new voice conversion system using speech synthesis based on short-time Fourier transform (STFT) magnitude. The system can modify two physical characteristics: one is the average pitch frequency, and the other is the spectral envelope. To modify the average pitch frequency the sampling rate conversion and the time-scale conversion are used. To modify the spectral envelope to the desired one the STFT magnitude is modified. Then, the output speech is synthesized using the modified STFT magnitude. The experiments show that the proposed system can modify these characteristics properly.