IMPLICIT PATCHING FOR DATAFLOW-BASED AUDIO ANALYSIS AND SYNTHESIS
Stuart Bray, George Tzanetakis · University of Michigan Library Repository · 2005
Programming software for audio analysis and synthesis is challenging.Dataflow-based approaches provide a declarative specification of computation and result in efficient code.Most practitioners of computer music are familiar with some form of dataflow programming where audio applications are constructed by connecting components with "wires" that carry data.Examples include networks of unit generators in Music-V style languages and visual patches in Max/Msp or PD.Even though existing dataflowbased audio systems offer a concise conceptual model of signal computation, this model does have limitations.In many cases, these limitations are a consequence of the programmer having to explicitly specify connections between components.Two such limitations are the difficulty of handling spectral data and the need for fixed-size buffers between components.In this paper we introduce Implicit Patching (IP), a dataflow-based approach to audio analysis and synthesis that attempts to address these limitations.By extending dataflow semantics a large number of connections are automatically created and buffer sizes can be changed dynamically.We describe Marsyas-0.2,a software framework based on IP, and comment on the strengths and limitations of the proposed approach.