基于正弦加噪声模型的说话人转换方法  

A Voice Conversion Scheme Based on Sinusoidal Plus Noise Model

作　　者：夏菁[1] 尹俊勋[1] 黄建成 黄锋[1] 

机构地区：[1]华南理工大学电子信息学院,广东广州510641 [2]摩托罗拉中国研究中心,上海200041

出　　处：《电声技术》2005年第2期49-52,共4页Audio Engineering

摘　　要：提出一种基于正弦加噪声模型的说话人转换方法，着重讨论通过修改音素段内的声学参数实现说话人的转换。通过修改基音频率和共振峰结构，该方法合成的语音有效地模拟了目标说话人的特性。听力测试表明，转换后的语音和目标说话人的语音相似度达到78.8％。与经典的LPC方法的对比实验验证了该法在合成语音质量方面的优越性。A voice conversion approach with a sinusoidal plus noise model is introduced and a parametric conversion algorithm based on phoneme segments is discussed in this paper. The modification of both pitch and formant structure contributed greatly to reproducing the target speaker's characteristics. Listening tests show that the similarity between target speech and modified one reached 78.8%. Compared to classical LPC approach, the experiments prove the superiority of this approach in terms of speech quality.

关 键 词：说话人转换 正弦加噪声模型 音素 基音 共振峰 

分 类 号：TN912.33[电子电信—通信与信息系统]