【摘要】 语音信号窗函数具有减少频谱能量泄露的作用,针对传统的语音加窗函数旁瓣衰减速度慢,信号频谱能量泄露大,不利于说话人识别特征参数提取的缺点,采用一种汉明自卷积窗函数取代汉明窗函数对语音信号预处理。为了进一步提高说话人系统的识别率,文章提出一种基于汉明自卷积窗的的一阶、二阶差分梅尔倒谱系数(MFCC)改进的动态组合特征参数方法。用高斯混合模型进行仿真实验,实验结果证明,用该方法提取的特征参数运用于说话人识别系统,相比于传统的MFCC说话人识别系统,其识别率大大提高。更多还原

【Abstract】 Speech signal windowing function can reduce spectral energy leakage,focusing on the shortcoming of the side lobe decay of traditional speech windowing function and the large energy leakage of signal spectrum, which it is disadvantage to the feature extraction of speaker recognition,a new window function is adopted to preprocess to the speech signal based on Hamming autocorrelation window function instead of Hamming window.In order to improve the recognition rate of speaker system,A new improved method is proposed in the paper to combine the first and second order of Mel Cepstral Coefficient combination dynamic feature parameters based on Hamming autocorrelation window function.The experimental results show that the proposed method can get better he recognition rate in Gaussian Mixture Model than that of the traditional speaker system with MFCC characteristic parameters, which can be applied to the speaker recognition system.更多还原