Intelligent Tools for Multitrack Frequency and Dynamics Processing

Abstract

This research explores the possibility of reproducing mixing decisions of a skilled audio engineer with minimal human interaction that can improve the overall listening experience of musical mixtures, i.e., intelligent mixing. By producing a balanced mix automatically musician and mixing engineering can focus on their creativity while the productivity of music production is increased. We focus on the two essential aspects of such a system, frequency and dynamics. This thesis presents an intelligent strategy for multitrack frequency and dynamics processing that exploit the interdependence of input audio features, incorporates best practices in audio engineering, and driven by perceptual models and subjective criteria. The intelligent frequency processing research begins with a spectral characteristic analysis of commercial recordings, where we discover a consistent leaning towards a target equalization spectrum. A novel approach for automatically equalizing audio signals towards the observed target spectrum is then described and evaluated. We proceed to dynamics processing, and introduce an intelligent multitrack dynamic range compression algorithm, in which various audio features are proposed and validated to better describe the transient nature and spectral content of the signals. An experiment to investigate the human preference on dynamic processing is described to inform our choices of parameter automations. To provide a perceptual basis for the intelligent system, we evaluate existing perceptual models, and propose several masking metrics to quantify the masking behaviour within the multitrack mixture. Ultimately, we integrate previous research on auditory masking, frequency and dynamics processing, into one intelligent system of mix optimization that replicates the iterative process of human mixing. Within the system, we explore the relationship between equalization and dynamics processing, and propose a general frequency and dynamics processing framework. Various implementations of the intelligent system are explored and evaluated objectively and subjectively through listening experiments.