A Semantic Approach To Autonomous Mixing

In this work, we give an overview of current automatic mixing approaches, and note the absence of using high-level, semantic knowledge to inform mixing decisions. We then propose a system where processing parameters for music production are inferred from instrument-specific rules rather than low-level features. To this end, we compile a body of rules from practical mixing engineering literature, and assess the potential of a system using just these rules. We compare the system to human mixing engineers and traditional automatic mixing systems that only use extracted, low-level signal features. Its performance is discussed and future directions are established.