项目名称: 语音信号声纹信息成分的深层表达
项目编号: No.61273264
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 戴礼荣
作者单位: 中国科学技术大学
项目金额: 81万元
中文摘要: 语音信号不仅包含有语言内容主要强信息成分,还包含有声纹信息成分等多种非语言弱信息成分。如何对语音信号各特定的信息成分进行有效表达,特别是对特定非语言弱信息成分的有效表达,如声纹信息成分的有效表达,是语音信号与信息处理中尚待解决的重要研究问题,也是阻碍在生物信息公共安全等领域具重大应用价值的声纹识别、声音转换等技术进一步发展的关键问题。本项目基于神经科学研究领域提出的深层表达原理,研究可有效表达语音信号中的特定声纹信息成分的深层表达可计算模型,包括层次性组件结构、模型构建模式、模型参数优化方法和算法、高效模型训练方法等;建立一种通过自动学习获得对语音信号中特定声纹信息成分进行有效表达且具一定推广性的深层表达方法;并应用于声纹识别和声音转换,以期显著提升声纹识别和声音转换的性能。本项目研究不仅具重要实际意义,对促进一般意义信号的弱信息成分分析这一信号处理领域基础问题的研究也具重要意义。
中文关键词: 深层表达;声纹信息;声纹识别;声音转换;
英文摘要: Speech signal is composed of not only linguistic dominant information component,but also other various non-linguistic minor information components such as voiceprint information component.How to effectively represent the different specific infromation component of speech signal,especialy how to effectively represent various non-linguistic minor information components such as voiceprint information component,is an unresolved important research problem in the filed of speech signal and information processing,and also is a key problom which limits the speech technology improvements such as voiceprint recognition and voice conversion that are found wide and important applications in the filed of public biology information security.Based on the deep representation principle indicated by the neuroscience research,the project proposes to study computational deep representation modelling with the ability to effecitvely represent the specific voiceprint information component of speech signal,including the hierarchical component structure、model constructing mode、model parameter optimization method and algorithms、model training method ,etc., aims to develop a deep representattion method which is capable of automaticly learning the effecitve representation for the specific voiceprint information component with good genera
英文关键词: Deep representation;;Voiceprint information;Voiceprint recognition;Voice conversion;