SGER: Music Similarity Retrieval Using Power-Law Metrics

SGER：使用幂律度量的音乐相似度检索manaris Bill Z.College of Charleston该项目正在开发一个基于识别美学相似性的音乐搜索引擎。该引擎利用幂律度量来提取音乐理论和音乐作品的其他属性的统计比例（例如，音高，持续时间，音高距离，持续时间距离，旋律间隔，和声间隔，旋律双谱等）。引擎使用均方误差（MSE）方法搜索与输入部分在美学上相似的部分。使用古典音乐档案语料库（14,695 MIDI作品）进行了初步测试，并结合了来自其他风格（例如爵士，摇滚，乡村等）的500多个MIDI作品。第一年的努力已经证明了附加文件格式的功能，包括MP3，这是基于web的音乐语料库中使用的主要格式。评估和验证实验将继续与表明检索件美学相似性的计算结果进行比较。这项研究可能会改变互联网音乐经济和功能。

AbstractIIS - 0849499 SGER: Music Similarity Retrieval Using Power-Law MetricsManaris Bill Z.College of Charleston The project is developing a music search engine based on identifying aesthetic similarities. This engine is utilizing power-law metrics to extract statistical proportions of music-theoretic and other attributes of music pieces (e.g., Pitch, Duration, Pitch Distance, Duration Distance, Melodic Intervals, Harmonic Intervals, Melodic Bigrams, etc.). The engine searches for pieces that are aesthetically similar to the input piece using a mean squared error (MSE) approach. Preliminary testing has been done using the Classical Music Archives corpus (14,695 MIDI pieces), combined with 500+ MIDI pieces from other styles (e.g. Jazz, Rock, Country, etc.). A first year effort has demonstrated functionality for additional file formats, including MP3, the predominant format used in web-based music corpora. Assessment and validation experiments will be continued to compare to computational findings indicating aesthetic similarity of retrieved pieces. This research is potentially transformative to the internet music economy and functionality.