大词汇量连续语音声像情报处理系统框架

System Framework for Large Vocabulary Continuous Speech Multimedia Intelligence Processing

分析了声像情报的自动标注与检索的研究概况及其关键技术,提出了声像情报的自动化处理框架以及基于语音内容的情报服务方案,讨论了声像情报特征提取、索引和分析过程中采用的主要情报内容特征。该方案在普通话广播语音和电话语音等数据测试中,准确率均在92%以上。

The automatic annotation and the retrieval of multimedia intelligence require large vo- cabulary speech recognition, natural language processing and information retrieval technologies. The automatic annotation and the retrieval of multimedia intelligence are studied and key tech- niques are analyzed. Then, a framework for multimedia intelligence automatic annotation and service is proposed. The main characteristics of speech recognition, speech indexing and intelli- gence analysis are investigated. The news broadcast and telephone audio data for annotation and retrieval are tested. Testing results show that the accuracy rate of the method is over 92%.