| 语音转换是一项转变措辞人声响特点的技巧,是将源措辞人声响转化为具有目的措辞人特点信息声响的办法。语音转换是语音旌旗灯号处置范畴的一个较新的分支,触及旌旗灯号处置、声学说话学、人工智能、形式辨认和盘算机迷信等学科范畴,它的研究对语音剖析、语音编码、语音分解、语音加强和语音辨认等语音旌旗灯号处置范畴有主要的增进感化。语音转换研究语音模子中各特点参数的变更纪律,对语音参数的分解、语音编码技巧的提高、措辞人加密和模仿技巧的成长有侧重要的增进感化；同时它在片子、电视节目中的配音、数字化的文娱等多媒体偏向、医学范畴、刑侦及保密通讯等方面也有着普遍的应用。语音转换技巧是对措辞人辨认和语音分解技巧的丰硕和延拓,有着优越技巧成长远景。是以语音转换技巧的研究具有辽阔的应用远景和主要的实际研究和适用价值。本论文彩用线性猜测剖析系数波形-叠接分解法完成语音转换,是基于提取源语音和目的语音特点的线性猜测系数转化的线谱对频率,树立结合参数的高斯混杂模子练习法,采取最年夜希冀法估量结合矢量高斯混杂模子的参数来肯定转换规矩,据转换规矩将源语音转换为猜测语音,再将猜测语音经由过程波形叠接法分解出转化后具有目的措辞人特点的语音。个中,在语音旌旗灯号转换的处置中,须要将线性猜测系数与线谱对频率参数互相转换。本文彩用运用余弦函数特征改良的Chebyshev多项式求根法,将余弦函数转换为高次幂函数再停止搜刮求根,来完成语音特点的线性猜测系数与线谱对参数的转化。 Abstract: Voice conversion is a change in the voice characteristics of the speaker's skills, is the source of the sound into a person with the voice of the speaker characteristics of the sound approach. Speech is speech signal processing of a new branch, touch signal processing, acoustic learning to speak, artificial intelligence, in the form of identification and computer science and other academic areas, its research on speech analysis, speech coding, speech decomposition, speech enhancement and speech identification speech signal processing category have main stimulative effect. Voice conversion model for the study of speech in the characteristic parameters of the change of discipline, the parameters of speech decomposition, speech coding techniques improve, the wording of people encryption and imitation skills grow side important stimulative action; bias and its voice in the television and film, digital entertainment and multimedia, medical category, criminal investigation and secure communication also has a widespread use. Voice conversion skills are the language of human identification and speech decomposition skills of the rich and extension, has a superior skill growth prospects. Is the study of voice conversion skills with a broad vision and the use of the main practical research and application value. The color with linear speculation analysis of waveform factor - fold after decomposition method to complete voice conversion is based on the extraction of speech source and purpose of the phonetic features of linear guess the conversion coefficient of the line spectrum frequency, establish the parameters of the Gaussian mixture model training method mixed with, to take the biggest hope method is used to estimate the amount of combination of vector Gaussian mixed model parameters to determine the conversion rules, according to the conversion rules to convert the source speech for guessing speech, then guess sound by the process of waveform stack connection decomposition transformation with the wording of the objective characteristics of speech. Among them, in the speech signal conversion processing, will have linear prediction coefficient and LSF parameter transformation. The application of cosine function features to improve the method of Chebyshev polynomial root finding, the cosine function conversion for high-order power function to stop search root, to complete the phonetic features of the linear predictive coefficient and line of parameter transformation. 目录: 摘要 5-6 Abstract 6-7 第1章 绪论 10-18 1.1 语音处理技术的发展 10-11 1.1.1 语音编码 10 1.1.2 语音识别 10 1.1.3 语音合成 10-11 1.2 语音转换技术的发展 11-17 1.2.1 国内外探讨近况 12-13 1.2.2 语音转换技术的运用 13-15 1.2.3 语音转换系统的探讨 15-17 1.3 论文主要探讨内容及安排 17-18 第2章 语音信号的基本特性和转换模型 18-26 2.1 语音信号的基本特性 18-20 2.1.1 语音信号的产生模型 18-19 2.1.2 语音信号的说话人特征 19-20 2.2 语音信号的转换模型 20-24 2.2.1 语音转换的探讨措施 20 2.2.2 语音转换的原理模型 20-24 2.3 本章小结 24-26 第3章 语音转换的关键技术的系统略论 26-39 3.1 语音转换的运用 26 3.2 语音转换系统的略论 26-36 3.2.1 语音转换探讨的层次结构 26-28 3.2.2 语音转换系统的总体构成 28-30 3.2.3 语音转换技术实现 30-36 3.3 语音转换性能测试 36-38 3.3.1 主观性能评估 36-37 3.3.2 客观性能评估 37-38 3.4 本章小结 38-39 第4章 语音转换系统设计 39-65 4.1 总体设计 39 4.2 语音转换的实现过程 39-59 4.2.1 训练阶段 39-52 4.2.2 转换阶段 52-56 4.2.3 语音合成 56-59 4.3 语音转换性能测试 59-60 4.3.1 主观测试措施 59-60 4.3.2 客观测试措施 60 4.4 实验设计和实验结果测试 60-64 4.4.1 实验语音的采集 61 4.4.2 实验系统的设计 61-62 4.4.3 实验结果的测试 62-64 4.5 本章小结 64-65 结论 65-67 参考文献 67-72 攻读硕士学位期间发表的论文和取得的科研成果 72-73 致谢 73 |
