摘要
通过提取基音频率、明亮度、带宽、过零率、响度、均方根、相邻点之间距离的均值和方差及Mel倒谱系数这8个特征构造特征集,在此基础上提出一种基于最近特征线的音频分类算法,对其进行枪声、鞭炮声、喇叭声及说话声的分类实验中,结果表明,该算法的分类效果较好,错误率可低至11.76%。
This paper constructs the feature set by extracting eight features including perceptual features like pitch frequency, brightness, bandwidth, zero-crossing rate, loudness, Root Mean Square(RMS), the distance between the adjacent point of the mean value and Mel Frequency Cepstral Coefficients(MFCC), and proposes an audio classification algorithm based on Nearest Feature Line(NFL). It is applied to classification experiment with four audio including guns, banger, horn and talks, and the result shows that the algorithm is effective in classification and its error rate can reduce to 11.76%.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第2期151-153,共3页
Computer Engineering
关键词
音频分类
最近特征线
音频特征选取
MEL倒谱系数
audio classification
Nearest Feature Line(NFL)
audio feature extraction
Mel Frequency Cepstral Coefficients(MFCC)