Rhythm analysis in the CompMusic project
In the CompMusic project we study five music traditions from an information retrieval perspective: Carnatic and Hindustani from India, Beijing Opera from China, Turkish-makam from Turkey, and Arab-Andalusian from the Maghreb. In this talk we will overview the computational rhythm analysis work done in the project, presenting various problems being addressed and methodologies being used. The music cultures under study share certain aspects at a broad level, such as syllable based percussion systems and well defined metrical structures, and they present several unique challenges that have not been very much addressed within the Music Information Retrieval field. Their computational analysis requires to develop specific approaches for each culture while extending the methodologies that are currently being used in MIR for rhythm analysis. In CompMusic we aim to build computational tools with which to explore large audio music collections of each culture by using features extracted from audio together with relevant contextual information. The tools developed are being integrated into Dunya (http://dunya.compmusic.upf.edu) and they should promote an enriched listening experience while enabling users to browse through music collections using audio content based similarity measures – similarity based on rhythmic content being one of them. We hypothesize that rhythmic similarity has contribution from rhythmic and metrical structures as well as from rhythm and percussion patterns, topics which are the focus of our work.
An important MIR rhythm task is the tracking and estimation of metrical structures, also referred to as meter inference. Useful in itself, it also provides a basis for further computational analysis such as structural analysis and extraction of rhythmic patterns. Though these metrical structures are well defined in these music cultures, the challenge arises from the flexibility and the scale of these structures. In the presentation we will formulate the problem of tracking the tala (the metrical framework of Carnatic music) from audio recordings and present the methodologies that we have used for the task. All the music repertories studied include percussion accompaniment. The percussion patterns played by the accompaniments are ubiquitous and often have important roles in defining many aspects of rhythm. Percussion patterns in Beijing Opera define and mark structural points in an aria. Percussion patterns in Indian art music are closely tied to the tala. However, the unique aspect of percussion in these cultures is the presence of a musically well motivated oral syllabic system, which provides a language for percussion. The syllables are an important part of music training and performance, and enable us to represent percussion patterns in a musically meaningful way. We will discuss how the syllabic system can be utilized in a wide variety of MIR tasks involving percussion patterns and present our recent work on percussion pattern transcription and classification in Beijing Opera. To carry out our data-driven computational analysis we have put together research corpora and specific annotated datasets. These corpora and datasets are public and will be available to the research community through the Dunya interface. In the presentation we will discuss the issues related to the rhythm annotated datasets that we have developed for meter inference tasks.
presentation slides available here