基于音频指纹的音乐检索方法优化研究  

Optimization of Music Retrieval Method Based on Audio Fingerprint

作　　者：何超宇 张继通 HE Chaoyu;ZHANG Jitong(Zhengzhou University of Industrial Technology,Zhengzhou 451100,China)

机构地区：[1]郑州工业应用技术学院,河南郑州451100

出　　处：《电声技术》2024年第9期133-135,共3页Audio Engineering

摘　　要：为研究一种高效的音乐检索方法,深入探讨音频指纹提取和并行检索优化。首先,采用谱聚类方法提取音频指纹,通过短时傅里叶变换(Short-Time Fourier Transform,STFT)和功率谱密度计算构建频谱特征矩阵,并利用谱聚类算法生成音频指纹向量。其次,利用Apache Lucene进行并行检索优化,通过分片索引和多线程处理显著提高检索效率。最后,以Freesound数据集为基础使用Librosa库实现特征提取,并通过Lucene进行并行检索实验。结果表明,基于音频指纹的音乐检索方法在检索时间方面显著优于常规策略。In order to study an efficient music retrieval method,audio fingerprint extraction and parallel retrieval optimization are discussed.Firstly,the audio fingerprint is extracted by spectral clustering method,the spectral feature matrix is constructed by Short-Time Fourier Transform(STFT)and power spectral density calculation,and the audio fingerprint vector is generated by spectral clustering algorithm.Secondly,Apache Lucene is used for parallel search optimization,and the efficiency of search is significantly improved through sharded index and multithreading.Finally,based on Freesound data set,Librosa library was used for feature extraction,and Lucene was used for parallel retrieval experiment.The results show that the music retrieval method based on audio fingerprint is significantly better than the conventional strategy in terms of retrieval time.

关 键 词：音乐检索 谱聚类 并行优化 音频指纹 

分 类 号：TN912.3[电子电信—通信与信息系统]