Unifying Low-Level and High-Level Music Similarity Measures

Abstract

Abstract-Measuring music similarity is essential for multimedia retrieval. For music items, this task can be regarded as obtaining a suitable distance measurement between songs defined on a certain feature space. In this paper, we propose three of such distance measures based on the audio content. First, a low-level measure based on tempo-related description. Second, a high-level semantic measure based on the inference of different musical dimensions by support vector machines. These dimensions include genre, culture, moods, instruments, rhythm, and tempo annotations. Third, a hybrid measure which combines the above-mentioned distance measures with two existing lowlevel measures: a Euclidean distance based on principal component analysis of timbral, temporal, and tonal descriptors, and a timbral distance based on single Gaussian MFCC modeling. We evaluate our proposed measures against a number of baseline measures. We do this objectively based on a comprehensive set of music collections, and subjectively based on listeners' ratings. Results show that the proposed methods achieve accuracies comparable to the baseline approaches in the case of the tempo and classifier-based measures. The highest accuracies are obtained by the hybrid distance. Furthermore, the proposed classifier-based approach opens up the possibility to explore distance measures that are based on semantic notions.

The State of the Art Ten Years After a State of the Art: Future Research in Music Information Retrieval

“…Hartmann [28] notes finding seven duplicates, but mentions no specifics. In [15,56,71,74], the authors describe GTZAN as having 993 or 999 excerpts. In personal communication, de los Santos mentions that they found seven corrupted files in Classical (though [71,74] report them being in Reggae).…”

“…In the 100 works using GTZAN, 96 employ the experimental design Classify [2-14, 16-18, 20-76, 78-100] (an excerpt is assigned a class, and that class is compared against a "ground truth"). In seven papers, GTZAN is used with the design Retrieve [15,19,22,38,77,79,101] (a query is used to find similar music, and the labels of the retrieved items are compared). The work in [8] uses GTZAN in the design Cluster (data is clustered, and the composition of the resulting clusters are inspected).…”

From Improved Auto-Taggers to Improved Music Similarity Measures

“…Another recent trend in the context of automatic tag classification is to use auto-tags for music similarity estimation [3,15]. The common main idea behind auto-tag based music similarity systems is to first estimate a song's tag profile and then compare the estimated tag profiles of two songs.…”

Hybrid retrieval approaches to geospatial music recommendation

“…Although quite a few MIR researchers suggest such a combination [2,1,3,12,5], a systematic evaluation of combining state-of-the-art audio and web similarity estimators is still missing, hence provided here.…”