SEMANTIC SIMILARITY MEASURE FOR MUSIC RECOMMENDATION A Pharos whitepaper (original) (raw)

From low-level to high-level: Comparative study of music similarity measures

Multimedia, 2009. ISM' …, 2009

Studying the ways to recommend music to a user is a central task within the music information research community. From a content-based point of view, this task can be regarded as obtaining a suitable distance measurement between songs defined on a certain feature space. We propose two such distance measures. First, a low-level measure based on tempo-related aspects, and second, a highlevel semantic measure based on regression by support vector machines of different groups of musical dimensions such as genre and culture, moods and instruments, or rhythm and tempo. We evaluate these distance measures against a number of state-of-the-art measures objectively, based on 17 ground truth musical collections, and subjectively, based on 12 listeners' ratings. Results show that, in spite of being conceptually different, the proposed methods achieve comparable or even higher performance than the considered baseline approaches. Furthermore, they open up the possibility to explore distance metrics that are based on truly semantic notions.

Unifying low-level and high-level music similarity measures

… IEEE Transactions on, 2011

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 low-level measures: a Euclidean distance based on principal component analysis of timbral, temporal, and tonal descriptors, and a timbral distance based on single Gaussian Mel-frequency cepstral coefficient (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.

Adapting Metrics for Music Similarity Using Comparative Ratings

2011

Understanding how we relate and compare pieces of music has been a topic of great interest in musicology as well as for business applications, such as music recommender systems. The way music is compared seems to vary among both individuals and cultures. Adapting a generic model to user ratings is useful for personalisation and can help to better understand such differences. This paper presents an approach to use machine learning techniques for analysing user data that specifies song similarity. We explore the potential for learning generalisable similarity measures with two stateof-the-art algorithms for learning metrics. We use the audio clips and user ratings in the MagnaTagATune dataset, enriched with genre annotations from the Magnatune label.

Incorporating machine-learning into music similarity estimation

2006

Music is a complex form of communication in which both artists and cultures express their ideas and identity. When we listen to music we do not simply perceive the acoustics of the sound in a temporal pattern, but also its relationship to other sounds, songs, artists, cultures and emotions. Owing to the complex, culturally-defined distribution of acoustic and temporal patterns amongst these relationships, it is unlikely that a general audio similarity metric will be suitable as a music similarity metric. Hence, we are unlikely to be able to emulate human perception of the similarity of songs without making reference to some historical or cultural context.

Learning music similarity from relative user ratings

Information Retrieval, 2013

Computational modelling of music similarity is an increasingly important task for personalisation and optimisation in Music Information Retrieval and for research in music perception and cognition. Relative similarity ratings provide a new and promising approach to this task as they avoid problems associated with absolute ratings. In this article, we use relative ratings from the MagnaTagATune dataset to develop a complete learning and evaluation process with state-of-theart algorithms and provide the first comprehensive and rigorous evaluation of this approach. We compare different high and low level audio features, genre data, dimensionality effects, weighted similarity ratings, and different sampling methods. For model adaptation, we compare SVM-based metric learning, Metric-Learningto-Rank (MLR), including a diagonal and a novel weighted MLR variant, and similarity learning with Neural Networks. Our results show that music similarity measures learnt on relative ratings are significantly better than a standard metric, depending on the choice of learning algorithm, feature set and application scenario. We implemented a testing framework in Matlab R , which we made publicly available 1 to ensure reproducibility of our results. 2

Algorithmic Prediction of Inter-song Similarity in Western Popular Music

Journal of New Music …, 2013

ABSTRACT We investigate a method for automatic extraction of inter-song similarity for songs selected from several genres of Western popular music. The specific purpose of this approach is to evaluate the predictive power of different feature extraction sets based on human perception of music similarity and to develop an algorithm able to reproduce and predict human ratings. The algorithm is a linear model that was trained and tested using perceptual data. We use publicly available algorithms to extract acoustic feature values from 78 songs used in a previous perceptual experiment. Feature value differences between songs are used in a multivariate linear regression calculation to find the optimal weighting coefficients for the feature values to best approximate the human similarity perception data. We use two evaluation methods: metrical and ordinal. We use a bootstrapping approach by randomly separating the experimental data into training and testing sets. We compare the performance of this model against the G1C model by Pampalk, winner of the MIREX 2006 competition on music similarity prediction. Both models produce a rather low performance on the metrical evaluation. However, on the ordinal evaluation, the linear regression model shows encouraging results (significantly outperforming the G1C algorithm): in the triadic comparison task, it can correctly predict 52.3 ± 0.5% of the most similar pairs, while the estimated theoretical maximum, based on participant consistency on the most similar pair rankings is 78 ± 8%. In a comparison of feature sets, we found the MIR toolbox to produce the best performance.

Improvements of audio-based music similarity and genre classification

proc. ISMIR, 2005

Audio-based music similarity measures can be used to au-tomatically generate playlists or recommendations. In this paper the similarity measure that won the ISMIR'04 genre classification contest is reviewed. In addition, further im-provements are presented. In particular, two ...

AL arge-Scale Evaluation of Acoustic and Subjective Music Similarity Measures

2004

Subjective similarity between musical pieces and artists is an elusive concept, but one that must be pursued in support of applications to provide automatic organization of large music collections. In this paper, we examine both acoustic and subjective approaches for calculating similarity between artists, comparing their performance on a common database of 400 popular artists. Specifically, we evaluate acoustic techniques based on Mel-frequency cepstral coefficients and an intermediate 'anchor space' of genre classification, and subjective techniques which use data from The All Music Guide, from a survey, from playlists and personal collections, and from web-text mining.

Content-based music recommendation based on user preference examples

1st Workshop On Music …, 2010

Recommending relevant and novel music to a user is one of the central applied problems in music information research. In the present work we propose three content-based approaches to this task. Starting from an explicit set of music tracks provided by the user as evidence of his/her music preferences, we infer high-level semantic descriptors, covering different musical facets, such as genre, culture, moods, instruments, rhythm, and tempo. On this basis, two of the proposed approaches employ a semantic music similarity measure to generate recommendations. The third approach creates a probabilistic model of the user's preference in the semantic domain. We evaluate these approaches against two recommenders using state-of-the-art timbral features, and two contextual baselines, one exploiting simple genre categories, the other using similarity information obtained from collaborative filtering. We conduct a listening experiment to assess familiarity, liking and further listening intentions for the provided recommendations. According to the obtained results, we found our semantic approaches to outperform the low-level timbral baselines together with the genre-based recommender. Though the proposed approaches could not reach a performance comparable to the involved collaborative filtering system, they yielded acceptable results in terms of successful novel recommendations. We conclude that the proposed semantic approaches are suitable for music discovery especially in the long tail.

Music Recommendation System Based on Artist Relatedness and Audio Similarity G . A . Vida

2019

The automatic music recommendation system has become an increasingly relevant problem in recent years, along with the increasing amount of music circulating in digital format.In this research, music recommendations are sought by searching for music that is similar to music input, using value of music features with K-Nearest Neighbor method. Artist relatedness also be used in this research to get music recommendations, so that the recommendations are suitable with the listener’s preferences. Spotify API which is provided by Spotify, an online music platform is used in searching music features and artist relatedness in this research. The method used to calculate audio similarity is K-Nearest Neighbor (K-NN). Based on evaluation result, music recommendations that only use artist relatedness features have a higher precision value compared to music recommendations that use combination of artist relatedness and audio similarity, because the research participants were more likely (subjecti...