Community Research and Development Information Service - CORDIS

Periodic Report Summary 1 - STAMUS (Statistical models for musical signal processing)

The objective of the STAMUS project is the development of innovative technologies for enabling access to the immense amount of audio music collections that are available on the Web. The STAMUS project seeks to build efficient models that allow the automatic extraction of musical content information from audio signals. Although there have been considerable advances in music storage, distribution, indexation and many other directions in the least decades, there are still some bottlenecks for the analysis and extraction of content information. Music audio signals are very complex, both because of the intrinsic nature of audio, and because of the information they convey. Often regarded as an innate human ability, the analysis of music and the automatic estimation of music content information proves to be a highly complex task, for at least two reasons. On the one hand, music signals are extremely rich and complex from a physical point of view, in particular because of the many modes of sound production, of the wide range of possible combinations between acoustic events, and also because signal observations are generally incomplete and noisy. On the other hand, music audio signals are also complex from a semantic point of view: they convey multi-faceted and strongly interrelated information such as harmony, melody, metric, and structure.

The STAMUS project aims at addressing two important limitations of state-of-the-art Music Information Retrieval (MIR) systems for estimating musical content from audio signals. At the statistical modeling level, existing models are limited in handling the semantic complexity of music. Also the ‘adverse environment’ signal features are necessarily corrupted by need to be taken into account. The STAMUS project has explored Markov Logic Networks (MLNs) as a highly flexible and expressive formalism for the harmonic analysis of audio signals. In this framework, models capable of handling both uncertainty and complex relational structure at multiple levels of representation have been developed. A deep understanding of the relationship between MLNs and probabilistic and logic-based models that are classically used in MIR has been provided, showing that this new framework opens interesting perspectives for the processing of (music) audio signals. The STAMUS project has also explored various structured signal representations that are useful for revealing and processing the various components that are usually superimposed in a music signal (tonal components, transients, background noise, etc.). The use of such well-structured representations allows providing access to higher-level information about the audio signal, taking into account the high variability of audio and the effect of the adverse environement. Finally during the STAMUS project, progress has been made toward the prospect of formalizing the process of human cognition in music and having a better understanding of the ability of human beings to obtain a sense of how music is organized.

The work developed within STAMUS project opens new perspectives towards the analysis of semantically complex music signals and to the development of multimodal approaches for music analysis. It participates to the continuous effort in the Music Information Retrieval community toward analyzing and formalizing the intricate mechanisms of music, as well as developing methods to enable or improve multimedia retrieval.

Contact

Véronique DEBISSCHOP, (Déléguée Régionale)
Tel.: +330169823294
E-mail

Subjects

Life Sciences
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top