Compositional Hierarchical Model for MIR

We developed a new hierarchical approach in a form of a compositional hierarchical model for music information retrieval. The model possesses the ability of unsupervised learning of features and a multi-layer structure of compositions, connected between the layers. Its main features are generative ability and transparency, which allow clear insight into the concepts learned from the input music signals.

The model consists of multiple layers, each is composed of a number of parts. The hierarchical nature of the model corresponds well with the hierarchical structures in music. Parts in lower layers correspond to low-level concepts (e.g. tone partials), while parts in higher layers combine lower-level representations into more complex concepts (tones, chords). The layers are learned in an unsupervised manner one-by-one from music signals. Parts in each layer are compositions of parts from previous layers based on statistical co-occurrences as the driving force of the learning process. We present the model's structure and compare it to other hierarchical architecture approaches in music information retrieval.

Spectral domain

The model was applied to spectral domain and used in automated chord estimation and multiple fundamental frequency estimation tasks.

Symbolic domain

A variation of the model named SymCHM and SymCHMMerge was also applied to the symbolic domain and used for the pattern discovery. The SymCHM and SymCHMMerge were also evaluated for the Discovery of Repeated Themes & Sections task at Mirex 2015 (see results) and Mirex 2016 (see results).

References

• M. Pesek, A. Leonardis, and M. Marolt, "Pattern discovery and music similarity with compositional hierarchical model," in CogMIR : August 12, 2016, Columbia University, New York, 2016, p. 8.
  [Bibtex]

  @conference{1537063619,
  author={Matevž Pesek and Aleš Leonardis and Matija Marolt},
  year={2016},
  pages={8},
  title={Pattern discovery and music similarity with compositional hierarchical model},
  booktitle={CogMIR : August 12, 2016, Columbia University, New York},
  }

• M. Pesek, A. Leonardis, and M. Marolt, "A compositional hierarchical model for music information retrieval," in Proceedings of the 15th Conference of the International Society for Music Information Retrieval, ISMIR 2014, October 27-31, 2014, Taipei, Taiwan, 2014, pp. 131-136.
  [Bibtex]

  @conference{1536035011,
  author={Matevž Pesek and Aleš Leonardis and Matija Marolt},
  year={2014},
  pages={131-136},
  title={A compositional hierarchical model for music information retrieval},
  booktitle={Proceedings of the 15th Conference of the International Society for Music Information Retrieval, ISMIR 2014, October 27-31, 2014, Taipei, Taiwan},
  }

Compositional Hierarchical Model applied to spectral domain