基于加权混合特征的话者识别算法

发布时间：2018-05-07 08:22

本文选题：多窗谱估计 + 伽马通滤波器组　；参考：《浙江工业大学学报》2017年06期

【摘要】：用多窗谱估计和伽马通滤波改进经典的梅尔倒谱特征(MFCC)的识别性能,并与delta特征相结合,提出了一种基于加权参数设置策略的混合特征话者识别算法.该算法解决了梅尔倒谱系数方差过大、听觉特征不明显及话者识别算法特征单一的问题.仿真结果表明:与MFCC和线性预测的提取方法相比,该算法鲁棒性能更优,对不同噪声环境的适应性更好.
[Abstract]:In this paper , the recognition performance of classical Mel cepstrum feature ( MFCC ) is improved by multi - window spectral estimation and gamma filtering and combined with delta feature . The algorithm solves the problem that the variance of the Mel cepstrum coefficient is too large , the auditory characteristic is not obvious and the speech recognition algorithm is single . The simulation results show that the robustness of the algorithm is better than that of MFCC and linear prediction , and the adaptability of the algorithm is better for different noise environments .

【作者单位】：浙江工业大学信息工程学院;
【基金】：国家自然科学基金资助项目(61471322,61402416)
【分类号】：TN912.3

【相似文献】