基于关键事件检测及信息熵计算的音频场景分割
首发时间:2012-11-14
摘要:在处理多媒体数字信息时,音频信息常常会起到关键的作用,特别是对多媒体文件进行语义层的解析、归纳和建索引时,更是如此。在一个连续音频流中,对关键事件检测,自动的对复杂音频进行语义层的分割和识别,在对音频文件基于内容的计算和分析中,起到了关键的作用。目前,很多研究工作都围绕对纯净、单一的音频事件进行检测和识别,而对于多媒体文件,如电影等,音频事件种类繁多,场景复杂,在一个具有完整语义信息的音频场景中,音频事件往往同时或交替出现。此时,现有的对单一音频事件的检测和识别就有很大的局限性。针对以上问题,提出了一个以音频事件为基本单位,通过对音频信息熵的计算和关键事件的检测,对音频流进行场景分割的方法。针对混合音频场景,把每一个音频流看做由时间上有序的音频事件组成,通过信息熵判决及音频事件检测(IEC-ED)确定每一个音频场景的端点,作为场景分割的依据。
For information in English, please click here
Audio Scene Segmentation Based on Information Entropy and Key Audio Event Detection
Abstract:Key audio information always plays a critical role in high-level multi-media information analysis, index and retrieval. Based on key audio event detection and audio scene segmentation in an input audio stream, high-level semantic reference can be automatically carried out in various content-based analysis applications. However, most previous researches are focused on the detection and classification of single audio events, as to that of multi-media information, such as movies which consist of relatively complicated content, are not efficient. When processing a complex audio scene where several key audio events occur simultaneously, the existing methods shows obvious limitations. In this paper, we propose a novel audio scene segmentation algorithm based on the calculation of information entropy and the detection of key audio events in audio signals. The end points of audio scenes can be determined by IEC-ED.
Keywords: Pattern Recognition Audio scene Audio scene segmentation Audio scene recognition
基金:
论文图表:
引用
No.****
同行评议
共计0人参与
勘误表
基于关键事件检测及信息熵计算的音频场景分割
评论
全部评论0/1000