基于自适应Mel滤波器组的MFCC特征提取的SOC设计-《郑州大学学报(工学版)》

文章信息/Info

Title:: Design of MFCC Feature Extraction Based on Adaptive Mel Filter Banks for SOC Appliacation

作者:: 陶佰睿; 郭琴; 苗凤娟; 李青龙; 1.齐齐哈尔大学通信与电子工程学院，黑龙江齐齐哈尔161006;2.中国科学院上海技术物理研究所，上海200083

Author(s):: TAO Bairui1; 2; GUo Qin1; MIAO Fengjuan1; 2; LI Qinglong1; 1.Computing Center,Qiqihar Universit,Qiqihar 161006,Chinma; 2. National Laboratory for Infrared Physics,Shanghai Insti-tute of Technical Physics,Shanghai 200083 ,China)

Keywords:: voiceprint authentication; adaptive mel filter banks; gender recognition; SOC

摘要:: 说话人声纹身份认证技术中的关键是特征参数的准确性和模式识别的速率.为此,对识别对象的性别予以区分,并进行参数可自适应调整的Mel滤波器组设计,即通过(Quartus ll平台在Altera 的 DE2系列型号为EP2C35F672C6的开发板上完成高效率说话人声纹特征提取的SOC(片上系统)设计.设计具体步骤如下:首先,设计截止频率为400 Hz和200 Hz的低通滤波器以完成男女生基音频率的检测;然后,依据计算出的每一帧语音频谱的频率范围确定Mel滤波器组的最高频率并完成参数设计;最后,在Quartus ll平台上完成Verilog-HDL代码设计,并封装为IP核完成SOC设计以及编译、仿真和下载验证.结果表明,Mel滤波器组利用率的提高有利于提高特征参数的准确性和识别速度.

Abstract:: The accuracy of characteristic parameter and pattern recognition rate among speaker voiceprint au-thentication technologies are important. In this paper, adaptive Mel filter banks are designed after the recogni-tion of the gender, and the SOC ( system-on-chip )design of high efficiency speaker voiceprint feature extrac-tion is completed on the EP2C35F672C6 development board of Alteras DE2 series. First of all，two low-passfilters cutoff frequency of 200 Hz and 400 Hz are designed to complete the pitch frequency detection of maleand female students. Then, the parameters of Mel filter banks are calculated by the highest frequency deducedfrom the frequency range of speech spectrum. Then, Verilog-HDL code encapsulated as lP core for SOC de-sign, compilation, simulation, and download authentication are finished on the Quartus ll platform. The re-sults show that adaptive Mel filter banks can improve both the accuracy of characteristic parameters and thespeed of recognition.