Feature design for intelligent control of the dynamic range compressor using audio decomposition
View/ Open
Publisher URL
DOI
10.26494/DMRN.2017.30583
Metadata
Show full item recordAbstract
We propose a method for the intelligent control of the dynamic range compressor targeting mono-timbral loops. Initial research using random forest regression has been shown to work in the context of isolated notes. Since audio loops have become the important in many production scenarios, this paper addresses this problem by decomposing loops into appropriate inputs for the initial system. We explore three types of audio decomposition approaches, onset event detection, NMF, and audio transient/stationary separation using ISTA, and extract features correspondingly. Results show a convincing trend that using features extracted in the decomposition domain to train the regression model improves the performance both numerically and perceptually.