用FPGA实现语音信号短时处理

目前,大多数语音识别系统都采用模式匹配的原理.根据这一原理,语音信号在预处理之后要进行特征提取,常用的特征包括短时平均能量或幅度、短时平均过零率等.本文围绕如何用FPGA实现语音信号常用特征提取,介绍了信号短时处理的概念和Altera的DSP Builder开发工具,并以短时能量和短时平均过零率为例,介绍了具体实现方法,给出了仿真结果,在EP1K100上验证了方法的正确性.实验证明,采用本系统可以有效地检测出语音信号的起始点和终点,从而为进一步语音处理,如语音合成、语音识别打下基础.

Short-ime processing of digital speech signal with FPGA

The methods of digital speech signal short-time processing with FPGA are presented in this paper.Firstly,the definition of short-time process and the development tool DSP Builder are introduced.Then,taking short-time energy and short-time zero crossing as examples,the implementation methods is presented in detail and the simulation results are given.It has been tested on the EP1K100 board.The start and the end of the speech signal can be detected successfully with the methods supplied by this paper.The result lays the foundation for the next work,such as speech synthesis and speech recognition.