基于共振峰和梅尔倒谱的声纹特征提取SOC设计

Voiceprint Feature Extraction SOC Design Based on Formant and Mel Cepstral

反映声道(谐振器)物理特性且不易受环境影响的元音共振峰可以更好地反映说话人的声纹特征,为此提出了说话人共振峰自适应MFCC(梅尔倒谱系数)特征提取SOC(片上系统)设计。首先提取说话人语音元音的三组共振峰来设计Mel三角形滤波器组,并基于传统MFCC与共振峰改进MFCC矩阵参数比值设计自适应融合说话人语音特征以改进MFCC。在MATLAB中完成性能仿真,在QUARTUS II中完成VERILOG-HDL代码设计,在FPGA(现场可编程门阵列)开发板上完成SOC设计、编译、仿真和验证下载。结果表明,在较高信噪比环境下,基于自适应融合和共振峰改进的MFCC得到的特征向量比传统的MFCC具有更强的鲁棒性,此技术在说话人声纹身份识别传感器设计中有较大应用推广价值。

Vowel formants,which reflect the physical characteristics of the vocal tract(resonator)and are not easily affected by the environment,can better reflect the speaker’s voiceprint characteristics.For this reason,the speaker formant adaptive MFCC(Mel Cepstral Coefficient)feature extraction SOC(system on chip)is proposed.Firstly,the three groups of formants of the speaker’s voice vowels are extracted to design the Mel triangular filter bank,and the MFCC is improved through adaptive fusion of the speaker’s speech features based on traditional MFCC and the formant improved MFCC matrix parameter ratio.The performance simulation is completed by using MATLAB and the VERILOG-HDL code design is completed by using QUARTUS II.Finally,the SOC design,compilation,simulation and verification download are completed on the FPGA(Field Programmable Gate Array)development board.The results show that the eigenvectors obtained by using the improved MFCC based on adaptive integration and formant are more robust than those got by using the traditional MFCC under the environment of high signal-to-noise ratio.The proposed technique has a great application promotion value in the design of speaker voiceprint recognition sensor.