A study on voice conversion with its application to CELP coder

• Main Author: 翁正平
• Other Authors: 簡福榮
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/64143139390334949700

碩士 === 國立臺北科技大學 === 電腦通訊與控制研究所 === 88 === In this thesis, a voice conversion system is developed to modify the speech signal of one speaker so that it sounds like that of another. Generally, two most important acoustic parameters are adopted for speaker adaptation. One is the pitch delay and the other is the short-time spectrum. In the former, we compare the PSOLA technique and propose a new and much better algorithm for pitch modification without suffering the problem of phase discontinuity. In the latter, the vocal tract areas are found to be efficient in representing the change of short-time spectrum information. In addition, the voice conversion system is built inside a CELP-based speech coder for the purpose of speaker security and low bit-rate requirement in communication networks. The result shows that the integrated system can easily convert the voice speech from one person to that of another unknown person by tuning the acoustic parameters mentioned above. It still takes advantage of noise suppression, while allowing additional benefits from reducing the unnatural impulse train due to voice conversion. Finally, the subjective MOS test is performed to measure the quality.