摘要
本文提出了一种基于加权Mel滤波器组的声学特征提取方法。该方法通过提取音频信号中的共振峰信息,使用动态自适应方法对中高频部分的Mel滤波器组进行加权,从而模仿人耳覆膜的频率选择映射。相比较于传统的MFCC特征,更适用于行车噪声环境下的快速声学事件检测任务;弥补了传统的Mel滤波器组高频部分分辨率低,从而导致对噪声鲁棒性较差的问题。实验结果表明:在信噪比较低的行车环境中,该特征有助于提高声学事件的检出率。
This paper presents an acoustic feature extraction approach based on weighting Mel filter banks. By extracting the formant information of audio signal,this method uses dynamic adaptive method for weighting high- frequency part of the Mel filter bank to achieve the purpose of simulating the auditory frequency selection mapping. Compared to the conventional MFCC feature,the proposed feature is more suitable for the fast acoustic event detection under driving noise task and makes up the poor robustness of the traditional methods which is resulting with the low resolution of Mel filter bank at the high-frequency part. The experiments show that the proposed feature helps to improve the detection rate of acoustic events in the low SNR driving environment.
出处
《智能计算机与应用》
2015年第3期16-18,共3页
Intelligent Computer and Applications
基金
国家自然科学基金(91120303)
关键词
声学事件检测
鲁棒性特征提取
行车噪声环境
动态自适应
MFCC
Acoustic Event Detection
Robust Acoustic Feature Extraction
Driving Noisy Environments
Dynamic Adaptive
MFCC