A system for automatic transcription of music based on multiple-agents architecture

This article presents an architecture of a system for automated transcription of solo piano music. The final goal is to build a system that will correctly transcribe digital recordings of polyphonic piano music without any additional end-user interaction. The transcription consists of two basic consecutive steps. The first step carries out time-frequency analysis and can be related to low-level auditory perception mechanisms that occur in the human ear. The second step implements perception mechanisms of note recognition and can be related to cognitive processes that occur in the brain. This stage is based on an emergent multiple-agent architecture, which combines top-down and bottom-up processing and is suitable for ill-posed problems.