短时频谱通用背景模型群联合韵律的年龄语音转换

发布时间：2018-01-11 21:09

本文关键词：短时频谱通用背景模型群联合韵律的年龄语音转换　出处：《声学学报》2017年06期 　论文类型：期刊论文

【摘要】：提出一种短时频谱通用背景模型群与韵律参数相结合进行年龄语音转换的方法。谱参数转换方面,同一年龄段各说话者提取语音短时谱系数并建立高斯混合模型,然后依据语音特征相似性对说话者进行聚类,每一类训练一个通用背景模型,最终得到通用背景模型群和一组短时频谱转换函数。谱参数转换之后再对共振峰进一步微调。韵律参数转换方面,基频和语速分别建立单高斯和平均时长率模型来推导转换函数。实验结果显示,提出的方法在ABX和MOS等评价指标上比传统的双线性法有明显的优势,相对单一通用背景模型法的对数似然度变化率提高了4%。这一结果表明提出的方法能够使转换语音具有良好目标倾向性的同时有较好的语音质量,性能较传统方法有明显提升。
[Abstract]:This paper presents a method of age speech conversion based on the combination of general background model group and prosodic parameters in short time spectrum. The speakers of the same age group extract the short-time spectral coefficients of speech and establish Gaussian mixture model, then cluster the speakers according to the similarity of speech features, and train a general background model for each class. Finally, the general background model group and a set of short-time spectrum conversion functions are obtained. After spectral parameter conversion, the resonance peak is further fine-tuned. The fundamental frequency model and the average time rate model are established to derive the conversion function. The experimental results show that the proposed method has obvious advantages over the traditional bilinear method in ABX and MOS. Compared with the single general background model, the logarithmic likelihood change rate of the proposed method is increased by 4. The results show that the proposed method can make the conversion speech with good target orientation and good speech quality. The performance is obviously improved compared with the traditional method.
【作者单位】：苏州大学电子信息学院;
【基金】：国家自然科学基金项目(61271360)资助
【分类号】：TN912.3
【正文快照】： 引言年龄语音转换作为具有表现力的语音合成技术之一,是指在保持文本信息不变的情况下,改变语音中说话者的年龄信息,将具有源年龄段特征的语音转换为具有目标年龄段特征的语音。年龄语音转换一般都要求是非特定人的,即转换模型适用于某个年龄段的所有说话者,而非某个特定说话

【相似文献】