A Compositional Hierarchical Model for Music Information Retrieval

This paper presents a biologically-inspired compositional hierarchical model for MIR. The model can be treated as a deep learning model, and poses an alternative to deep architectures based on neural networks. Its main features are generativeness and transparency that allow clear insight into 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 lowerlevel representations into more complex concepts (tones, chords). The layers are unsupervisedly learned one-byone 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 deep architectures. A preliminary evaluation of the model’s usefulness for automated chord estimation and multiple fundamental frequency estimation tasks is provided. Additionally, we show how the model can be extended to event-based music processing, which is our final goal.