Ulises: A Agent-Based System For Timbre Classification

Abstract

The Sound and Music Computing (SMC) field has grown over the years and every time there are more conferences and specialized researchers in this area. The sub-field of Music Information Retrieval (MIR), one of the main research fields on SMC has focused on getting information from sound data. The most critical issue with regard to the human perception of sound is: what are the qualities of musical instrument sounds to perform recognition of its sound sources. There are four main sound dimensions: pitch, loudness, duration and timbre. The fourth dimension, timbre, is the most vague and complex dimension, a complex and high-level multidimensional property. Recognition of timbres is an area of high interest within MIR, being present in several papers state of the art on SMC. About Multi-Agent Systems (MAS), the term autonomous refers to the fact that the agents have their own existence, regardless of the existence of other agents, and are able to take own decisions without outside interference. Agents technology is particularly suitable for musical applications because of the possibility of associating a computational agent with the role of a singer or instrumentalist as can be seen in works state of art in SMC area. In this context, this paper proposes a agent-based approach to timbre recognition, focusing on the parallelization of the classification model. For this, we assign a method of recognition of timbres to different agents, where each agent is a specialized entity in a particular timbre, characteristic of a specific instrument, seeking a distributed solution for solving the timbre recognition problem.