实用语音识别的场景标记辅助系统

Multimedia scene labeling based on speech

标引是通过给音频-视频数据加入标记,对其内容进行描述,以便于信息的检索和查询。语音标引在媒体资产管理中扮演了很重要的脚色。介绍了一种基于EBF网络的语音标引辅助系统,该系统可自动识别标引员所说的短语,辅助标引员在视频媒体上实现标引。系统从语句中将这些短语分割出来,并通过EBF神经网络进行建模。实验结果证明,该系统具有实用性,在媒体资产管理方面有良好的应用前景。

The main objective of the indexing process is to assign labels to the audio-visual data in order to describe its content. Audio indexing plays a key role in this process. In this paper, a speech-based man-machine labeling system for media asset management is presented. The system recognizes the phrases spoken by the human annotator automatically and assists him to mark up shots of subjects in a video sequence. The phrases are segmented from short sentences and modeled by the elliptical basis function （EBF） networks. Experimental results indicate that the speech-based labeling system is practical and has great promise for media asset management.