基于MFCC参数的说话人特征提取算法的改进

An Efficient Speaker Feature Extraction Method Based on MFCC

在说话人识别系统中,特征参数的提取对语音训练和识别有着重要的影响。对于特征参数提取模块,提出了一种新的特征参数提取算法MFCC_E(Efficient MFCC)。相对于标准算法MFCC_S(Standard MFCC),MFCC_E在特征提取模块部分减少了53%的计算量。最终实验结果说明MFCC_E的识别率为90.3%,仅比标准MFCC算法92.0%的识别率降低1.7%。因为MFCC_E算法的这种特点,使其能够更有效的适用于硬件实现。

Feature extraction is a significant module for speech training and recognition in speech recognition system. A new algorithm of feature extraction MFCC E（Efficient MFCC） is introduced. Compared to the standard algorithm MFCC_S （Standard MFCC）, the new algorithm reduces the computation power by 53%. The simulation results indicate MFCCE has a recognition accuracy of 90.3%, and there is only an 1.7% reduction compared to MFCCS which has 92.0% recognition accuracy. The new algorithm is acceptable for hardware implement for its advantage.