Section: New Results

Music Content Processing and Music Information Retrieval

Acoustic modeling, non-negative matrix factorisation, music language modeling, music structure

Music structure modeling by System & Contrast

Participants : Frédéric Bimbot, Corentin Louboutin.

The System & Contrast (S&C) model aims at describing the inner organization of structural segments within music pieces in terms of : (i) a carrier system, i.e. a sequence of morphological elements forming a multi-dimensional network of self-deducible syntagmatic relationships and (ii) a contrast, i.e. a substitutive element, usually the last one, which partly departs from the logic implied by the rest of the system [16] .

With a primary focus on pop music, the S&C model provides a framework to describe internal implication patterns in musical segments by encoding similarities and relations between its constitutive elements so as to minimize the complexity of the resulting description. It is applicable at several timescales and to a wide variety of musical dimensions in a polymorphous way, therefore offering an attractive meta-description of different types of musical contents.

We have established the filiation of the S&C model as an extension of Narmour's Implication-Realization model [104] , [105] and Cognitive Rule-Mapping [103] .

We have introduced the Minimum Description Length scheme as a productive paradigm that supports the estimation of S&C descriptions and establishes promising connections between Music Data Processing and Information Retrieval on the one hand, and modern theories in Music Perception and Cognition on the other hand, together with interesting perspectives in other areas in Musicology.

The model is currently being investigated for the multi-scale description of chord sequences.

Tree-based representation of music pieces

Participants : Frédéric Bimbot, Corentin Guichaoua.

Modeling music structure, i.e. the organisation of musical elements and their relationships within a piece of music, is an open problem of primary importance in MIR.

To address this challenge, we approach music structure description as the inference of a low complexity generative grammar able to account for the music piece, itself represented as a sequence of symbols.

Originally introduced for the inference of structure in DNA sequences, Straight-Line Grammars (SLG) form a particular subclass of Context-Free Grammars (CFG) which can be used to model symbolic sequences and to represent them as hierarchical trees. However, SLGs appear to be poorly suited to some particularities of musical patterns, such as segmental regularities, closure substitutions and specific style structures.

We are designing and investigating formal and algorithmic extensions of SLGs as SLEGs (Straight-Line Edition Grammars). Based on a more general minimum description criterion, the SLEG extension allows alterations in the generation step and enables the use of priors in the grammar inference process. Current work includes a diagnostic comparison between the various approaches on the structural segmentation of chord sequences from pop songs.