Fig 1.
(a) Series corresponding to the 30 s sample of a string quartet by Claude Debussy, (b) local standard deviation series of the signal represented in (a).
Fig 2.
Cartesian representation of a series of eight points {Vj} (a) and their respective visibility graph (b).
Fig 3.
Visibility graph generated from Debussy string quartet {Vj} series (Fig 1).
Fig 4.
(a) Music signal with strong PIH (heavy metal style); (b) music signal with poor PIH (classical style).
Fig 5.
Local standard deviation series of 30 s audio samples.
The color blue represents the jazz genre, and red and black represent the hip-hop and classical genres, respectively.
Fig 6.
{Vj} Series (left) and their respective visibility graphs (right).
The colors in the graphs are the communities, which are obtained from modularity.
Table 1.
Mean and standard deviation of the topological properties of visibility graphs.
Fig 7.
Mean Q (a) and 〈k〉 (b) calculated from 100 visibility graphs labeled in 10 musical genres.
Fig 8.
Difference in the mean Nc between pairs of musical genres according to Tukey’s test.
Black boxes represent statistically significant differences, and white boxes represent nonsignificant differences.
Table 2.
Pairs of musical genres with significant differences for clusters formed with the ASVD components, according to Tukey’s test.
Table 3.
Classification using artificial neural networks and ASVD attributes.
Fig 9.
True positives rate for classification using neural networks, where the input attributes are the ASVD or beat histogram.
Fig 10.
Results of classification with neural network: Audio Signal Visibility Descriptor (ASVD) & Audio Signal Processing Descriptors (ASPD) compared with beat histogram & ASPD.
Fig 11.
Gain ratio for the attribute selection.
Fig 12.
Result of the classification of musical genres in the GTZAN database in our proposal and in the experiment of [7].
Fig 13.
SANN with a hidden layer of 16 neurons considering ASVD at the input layer.
Fig 14.
SANN with a hidden layer of 16 neurons considering BH at the input layer.
Fig 15.
SANN with two hidden layers of 32 neurons and ASVD + ASPD at the input layer.
Fig 16.
SANN with a hidden layer of 32 neurons and BH + ASPD at the input layer.