CAJ | 학술논문

提出一个基于乐纹特征和倒排索引的音乐检索系统。该系统由预处理、特征提取、索引和精匹配四部分组成。通过两次基于动态阈值的筛选，选取频谱中最为稳定的点作为特征点，将特征作为关键词，采用倒排索引实现系统的初次查询。而精匹配则是对初次查询的优化和重排序，用优化后的编辑距离来计算两个特征序列的相似度。实验结果表明，提取的特征数据较小，系统具有较高的鲁棒性和查询准确率。
제출일개기우악문특정화도배색인적음악검색계통。해계통유예처리、특정제취、색인화정필배사부분조성。통과량차기우동태역치적사선，선취빈보중최위은정적점작위특정점，장특정작위관건사，채용도배색인실현계통적초차사순。이정필배칙시대초차사순적우화화중배서，용우화후적편집거리래계산량개특정서렬적상사도。실험결과표명，제취적특정수거교소，계통구유교고적로봉성화사순준학솔。
This paper presents a music retrieval system,it is based on audio fingerprint and inverted index.The system consists of four parts:preprocessing,feature extraction,index,and precise matching.We select the most stable points in the spectrum as the feature points after two screening based on dynamic threshold,take the feature for keyword,and use inverted index to implement initial query of the system. And the precise matching is the optimisation and reordering on the initial query.We use the optimised Edit distance to calculate the similarity of the two feature sequences.Experimental results show that the feature data of audio features is very small and the system has high robustness and query precision.