摘要
在基于哼唱的音乐检索系统的研究中,对哼唱输入与数据库进行合理的近似匹配以及有效的检索方法的研究不断深入,但对于哼唱音频信息的有效处理以提取准确有效的旋律特征信息构造查询的研究还不充分。本文在已有的哼唱信息处理方法之上,提出了一种结合了哼唱语音信号增强技术以及时域与频域处理技术的哼唱转谱方法,包括采用了分级音符分割方法、基于规则的音高跟踪方法,并提出一种合理的旋律特征表达的中间格式,用于哼唱查询构造。实验结果证明了这种对哼唱信息转谱处理新方法的有效性和准确性。通过降低哼唱转谱过程中引入的误差,进而可以有效地提高整个音乐检索系统的性能。
More research in query-by-humming system has focused on the approximate matching approaches to search the database using sung queries. But the researches of effective humming query processing methods and exact melodic character extraction and representation have not been examined maturely. In this paper, an effective and novel humming query processing method has been presented, which combines the speech enhancement technic with tlme-domain and frequency-domain analysis to do the humming transcription. A hierarchical note segmentation method and a rule-based pitch tracking method have been taken and an intermediate format for melody representation has been proposed. Experimental results show the effectiveness and accuracy of the humming transcription, thus the performance of following melody based searching can be improved.
出处
《计算机科学》
CSCD
北大核心
2006年第5期169-172,190,共5页
Computer Science
基金
国家部委科技电子预研项目资助(413160501)
关键词
哼唱信息处理
音符分割
音高跟踪
旋律表示
数据库
Humming query processing, Note segmentation, Pitch tracking, Melody presentation
