语音信号倒谱特征提取建模与仿真

Modeling and Simulation of Extract Cepstrum Features of Speech Signal

随着嵌入式语音识别的应用形式和领域的不断拓展,系统设计的复杂性也随之增加,仅仅支持单一数学模型的系统级设计工具已经不能满足这类复杂系统设计的需求。为了在系统执行这个层面之上更精确定义提取语音信号特征的行为,研究了面向角色的系统级建模方法,分析和论述了如何利用线性预测分析方法提取语音倒谱特征参数的基本原理,介绍了嵌入式系统级建模和仿真平台PtolemyII,最后在PtolemyII平台上完成语音信号倒谱特征提取的建模,并对模型中的预处理、端点检测等进行了仿真实验。

With the development and increasing popularization of embedded speech recognition applications, the system design becomes more and more complex. For the heterogeneous characteristics of embedded systems, a platform with single model of computation can't do well in designing these systems. A significant part of design problem is how to use formal model and high-level synthesis to ensure safe and correct designs. The actor-oriented system-level model method was focused on and how to extract cepstrum feature parameters was studied by using linear predictive analysis method. In addition, the embedded system modeling and simulation platform Ptolemy II were described. Cepstrum features extract model was designed on Ptolemy II, and the simulation of preprocess, end point detection and LPC order choice was given.