摘要
将听觉感知特征用于声学目标识别,实现对稳态噪声样本的识别,证明听觉感知特征在声学目标识别领域中的有效性。首先通过语音和乐音识别领域的研究成果确定听觉感知特征的提取方法,然后在汽车、飞机、舰船和乐器等不同类型的声目标中进行识别研究。结果显示:对于平稳噪声样本,听觉感知特征中的谱特征具有突出的优越性,其中基于Mel倒谱系数、谱下降值、分段谱质心、分段谱质心带宽和分段谱通量能够识别目标,且具有较高的识别率和稳健性。
The efficiency of recognition features is one of the key problems in recognizing acoustic targets accurately. Simulating the function of auditory system proposed in recent years has been viewed as a new feature extraction approach. In this study, auditory perceptual features were applied into recognizing the steady-state sound, which was investigated to prove which feature is effective. Firstly, the extraction approaches of the auditory perceptual features were previously identified by researching on the musical acoustics and speech literature. Secondly, a number of recognition researches were performed to recognize the acoustic targets in different fields, such as vehicles, airplanes, ships and musical instruments. Finally, the recognition results show that the spectral signal features of auditory perceptual features are useful for steady-state sounds classification. The features which have good and steady performance include： the mel-Frequency cepstrum coefficients and the spectral roll-off, as well as several features based on sub-band of spectral centroid, bandwidth and flux.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第10期3128-3132,共5页
Journal of System Simulation
关键词
听觉感知特征
特征提取
目标识别
谱质心
auditory perceptual feature
feature extraction
target recognition
spectral centroid
