摘要
本文介绍从语音的有限个共振峰估计声道面积函数模型参数的扰动实现方法.在Schroeder的声管喇叭方程的扰动解法基础上,采用改进的声道面积函数模型;针对不同共振模式之间的交叉影响,采用了交叉敏感扰动矩阵关系式,及交叉扰动敏感矩阵的测试方法.设计了用递归式扰动法实现声道面积参数估计的算法.利用“共振峰─扰动”映射码书作为语音共振峰匹配最佳的初始面积扰动矢量,井用多级扰动目标和自适应增量控制,可有效地实现从有限个语音目标共振峰逆向估计声道面积函数.
A method for estimating the parmeters of the vocal-tract area function from the first finite formants frequencies of speech signal is propsoed. An improved vocal-tract area function model is suggested based on the perturbation approach to the solution of the Webster's wave equation of the acoustic resonance tube by Schroeter. By considering the cross-influence of the different resonance modes of the perturbation,a cross-sensitivity perturbation relation in matrix form is presented. A method to measure the cross-sensitivity matrix is proposed. An iterative perturbation algorithm for determining the area function parameters from formant targets is developed. By looking up a 'formant-to-perturbation 'mapping codebook,an optimal initial perturbation vector for given formant targets is selected.A step-by-step sub-target perturbation procedure and adaptive factor for perturbation increment are adopted.An effective realizaion of the estmation of the vocal-tract area function is illustrated with comparison to the published data.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1997年第7期32-37,共6页
Acta Electronica Sinica
关键词
语音信号处理
语音合成
低比特率
语音编码
Speech signal processing ,Speech synthesis, Low bit rate speech coding