arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2201.11178cs.SDcs.IReess.AS
收藏

快速搜索相似音频项的解决方案

Rapid solution for searching similar audio items

University of Ljubljana(卢布尔雅那大学)

Kastriot Kadriu

摘要:

用于找到相似音频项的朴素方法将是以k最近邻方式将来自测试示例的特征向量的每个条目与候选的每个特征向量进行比较。这种方法已经存在两个问题:音频信号由高维向量表示,并且候选的数量可能非常大-考虑数千个。搜索过程将非常复杂。我们的论文将通过散列方法来处理这个问题,更具体地说,是局部敏感散列。本项目将本着分类和聚类的精神对问题进行研究。计算机声音产生原理将用于确定描述音频信号的哪些特征最有用。这将降低特征向量的大小,并随后加速该过程。

英文摘要:

A naive approach for finding similar audio items would be to compare each entry from the feature vector of the test example with each feature vector of the candidates in a k-nearest neighbors fashion. There are already two problems with this approach: audio signals are represented by high dimensional vectors and the number of candidates can be very large - think thousands. The search process would have a high complexity. Our paper will treat this problem through hashing methodologies more specifically the Locality Sensitive Hashing. This project will be in the spirit of classification and clustering problems. The computer sound production principles will be used to determine which features that describe an audio signal are the most useful. That will down-sample the size of the feature vectors and speed up the process subsequently.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑