基于空间线索时域梯度的音频关注度计算模型  

Spatial Cues Gradient in Time Domain Based Audio Attention Computational Model

作　　者：杭波[1] 王毅[1] 康长青[1] 黄健[1] 

机构地区：[1]湖北文理学院数学与计算机科学学院,湖北襄阳441053

出　　处：《系统仿真学报》2016年第10期2369-2377,共9页Journal of System Simulation

基　　金：国家自然科学基金(61201247);湖北省自然科学基金(2011CDB322)

摘　　要：虚拟现实中方位快速变化的音频信号应当具有较高的关注度,但现有自底向上音频关注度计算模型提取底层音频特征如能量、基音、过零率等,无法有效表达该类信号引起的音频关注度,有可能造成漏检。针对此问题,基于空间信息对关注产生影响的心理学原理,引入空间线索短时变化梯度,用以度量单声源空间方位快速变化引起的关注。计算由各子带空间线索纽成的空间线索矢量的短对变化均值作为空间线索变化梯度,建立基于空间线索变化梯度的音频关注度模型。与当前音频关注度计算模型相比,关注音频的检出率提高了4.5个百分点。In virtual reality audio, sound source whose directions change rapidly should have higher attention level. But present bottom-up audio attention computational models extract the underlying characteristics of single channel audio such as energy, pitch, zero crossing rate etc., which can not effectively express the audio attention caused by such signals. To solve this problem, based on the psychological principles that spatial information affects attention, a model was proposed to introduce the short-term spatial gradient cues to measure the attention caused by the single audio source space direction changing. Compared to the traditional audio attention computational model, the recall of detection of attention audio events increased 4.5 percentage points in experiments.

关 键 词：音频 关注度计算模型 空间线索 梯度 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]