Automatic music genre classification using ensemble of classifiers

Silla, Carlos N.; Kaestner, Celso A. A.; Koerich, Alessandro L.

doi:10.1109/icsmc.2007.4414136

Cited by 44 publications

(36 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Despite the complex situation that emerges in the problem of automatic genre classification [41][42][43][44], our model is very simple. From the qualitative point of view, the characteristic of songs and music genres is related with multidimensional aspects like timbre, melody, harmony, rhythm, among others.…”

mentioning

confidence: 99%

Universal patterns in sound amplitudes of songs and music genres

et al. 2011

View full text Add to dashboard Cite

We report a statistical analysis over more than eight thousand songs. Specifically, we investigate the probability distribution of the normalized sound amplitudes. Our findings seems to suggest a universal form of distribution which presents a good agreement with a one-parameter stretched Gaussian. We also argue that this parameter can give information on music complexity, and consequently it goes towards classifying songs as well as music genres. Additionally, we present statistical evidences that correlation aspects of the songs are directly related with the non-Gaussian nature of their sound amplitude distributions. In recent years, studies of complex systems have become widespread among the scientific community, specially in the statistical physics one [1][2][3][4][5]. Many of these investigations deal with data records ordered in time or space (i.e., time series), trying to extract some features, patterns or laws that may be present in the systems studied. This approach has been successfully applied to a variety of fields, from physics and astronomy [6] Music is a well known worldwide social phenomenon linked to the human cognitive habits, modes of consciousness as well as historical developments [14]. In the direction of music's social role, some authors investigated collective listening habits. For instance, Lambiotte and Ausloos [15] analyzed data from people music library finding audience groups with the size distribution following a power law. They also investigated correlations among these music groups, reporting non-trivial relations [16]. In another work, Silva et al.[17] studied the network structure of the song writers and the singers of Brazilian popular music (mpb). There is also an interest in the behavior of music sales [18] as well as in the success of musicians [19][20][21].Despite these cultural aspects, songs form a highly organized system presenting very complex structures and long-range correlations. All these features have attracted the attention of statistical physicists. In a seminal paper, Voss and Clarke [22] analyzed the power spectrum of radio stations and observed a 1/f noise like pattern. They also showed that the correlation can extend to longer or shorter time scales, depending on the music genre. Hsü and Hsü[23] investigated the changes of acoustic frequency in Bach's and Mozart's compositions, finding self-similarity and fractals structures. In contrast, they report no resemblance to fractal geometry [24] for modern music. Fractal structures have also been reported in the study of sequences of music notes [25], where Su and Wu [26] suggest that the multifractal spectrum can be used to distinguish different styles of music. In this brief literature review, we see that special attention was paid to the fractal structures of music, correlations and power spectrum analysis. However, much less attention has been paid to the understanding of the amplitude distribution. This last point has been noted by Diodati and Piazza[34]. In their work, they investigated the distribution of t...

show abstract

mentioning

confidence: 99%

Universal patterns in sound amplitudes of songs and music genres

et al. 2011

View full text Add to dashboard Cite

show abstract

“…No mesmo ano, Kosina desenvolveu em [5] um sistema de classificação de gêneros musicais, chamado MUGRAT, obtendo aproximadamente 88% de reconhecimento para três gêneros musicais. Silla et al [11] realizaram uma análise diferente na qual, em vez de extraírem informações de toda a música, dividiram o áudio em três partes (começo, meio e fim, com trechos de 30 segundos cada), utilizando classificadores diferentes para cada parte e combinando os resultados. Nesse trabalho, foi alcançada uma taxa média de acerto 3% maior do que o melhor resultado obtido individualmente (55,15% sem a combinação contra 58,07%, utilizando três segmentos, para dez gêneros musicais).…”

Section: Trabalhos Correlatosunclassified

Reconhecimento automático de gêneros musicais utilizando classificadores baseados em múltiplas características

Romano¹,

Adami

2015

Revista Brasileira De Computação Aplicada

View full text Add to dashboard Cite

Resumo: O reconhecimento de gêneros musicais é um importante problema de pesquisa que tem recebido muita atenção de pesquisadores e profissionais de música. Os benefícios dessa área podem ser aplicados em diversas tarefas, como na organização de bases de áudios digitais, na construção de novos mecanismos de buscas e recomendações de músicas. Este artigo propõe um método para reconhecimento automático de gêneros musicais baseado na variação da arquitetura padrão dos sistemas de reconhecimento de gêneros musicais, envolvendo técnicas como extração por segmentos, sistemas multiexpert e combinação de classificadores (paramétricos e não paramétricos). O reconhecedor foi avaliado em uma base de gêneros musicais própria extraída de um website de recomendações musicais e na base GTZAN, com dez gêneros musicais cada, onde obteve uma taxa de reconhecimento de 46,2% e 75,1%, respectivamente.Palavras-chave: Processamento de sinais. Reconhecimento de gêneros musicais. Reconhecimento estatístico de padrões. Abstract: Automatic music genre recognition is an important research topic that has received much attention from researchers and digital music professionals. The benefits of such field can be applied to various tasks, such as in the organization of digital audio databases, in building new

show abstract

“…In this table we employ a binary BME mask -for (B)eginning, (M)iddle and (E)nd time segmentswhere 1 indicates that the feature was selected in the corresponding time segment, and 0 otherwise. [31], [33] and [34]. We remember that features 1 to 6 are Beat related, 7 to 25 are related to Timbral Texture, and 26 to 30 are Pitch related.…”

Section: Methodsmentioning

confidence: 99%

“…We use feature space decomposition following the OAA and RR approaches, and also features extracted from different time segments [31], [33], [34]. Therefore several feature vectors and component classifiers are used in each music part, and a combination procedure is employed to produce the final class label for the music.…”

Section: The Space-time Decomposition Approachmentioning

confidence: 99%