基于FPGA的Mel倒谱系数提取方法设计与实现

Design and Implementation of Mel-cepstral Coefficients Extraction Method Based on FPGA

为了在嵌入式系统中实现参数化语音合成,提出一种在FPGA(field programmable gate array)中通过SOPC(system on programmable chip)系统提取Mel倒谱系数的方法。用SOPC Builder搭建硬件平台,然后在嵌入式NiosⅡ处理器中对语音信号做短时傅里叶变换、对数幅度运算、逆傅里叶变换,最后通过牛顿迭代法实现Mel倒谱系数的求解。将NiosⅡ处理器得到的Mel倒谱系数通过JTAG接口回传到计算机上,并与MATLAB仿真结果进行比较,结果表明NiosⅡ处理器提取的Mel倒谱系数与MATLAB提取的系数结果相同。

A Mel-cepstral coefficients（MCCs） extraction method on FPGA（Field Programmable Gate Array） by using SOPC（System on a Programmable Chip） system for implementing parameter speech synthesis on embedded system was proposed.A hardware platform was established by SOPC Builder.Short Time Fourier Transform（STFT） was firstly performed on input speech signal.Then the results of STFT were transformed by a logarithm calculation.Finally the Mel-cepstral Coefficients（MCCs） were extracted by Newton-Raphson method from the results of inverse fast Fourier transform that is obtained on the results of logarithm transform.The MCCs calculated by Nios Ⅱ processor are then feedback to computer through JTAG interface for comparing with the simulating results of MATLAB.The results show that the extracted MCCs by Nios Ⅱ are same with the results of MATLAB.