Automatic mixing research

Tracking academic work in the field of automatic multitrack audio mixing

Click the buttons below to filter the table of papers.

Level Equalization Compression Panning Reverb Multiple Machine Learning Knowledge-based Overview Clear

Categories

Mixing is a complex process carried out by audio engineers during the music production process. It involves a number of different steps each of which oftentimes has many sub-steps dependent on the source material and the desired result. In the literature different systems have been proposed for these different processes. In order to better categorize the work in this area the following mix processes are identified:

  • Level - setting the relative level of the elements within a mix
  • Panning - placement of the elements within the stereo field
  • Equalization - adjustment of the timbre of elements within a mix
  • Compression - application of dynamic range compression
  • Reverb - application of artificial reverberation
Systems that incorporate more than one of these are labeled Multiple.

Approaches

Two main approaches1 in building automatic mixing systems have been addressed in the literature:

  • Knowledge-based systems (KBS) systems generally consist of a set of rules or guidelines collected from professionals or hand-crafted by the designer, as well as an algorithm for applying this knowledge for a task.
  • Machine Learning (ML) approaches differ in that they use algorithms and statistical models directly on data, generally through the application of some kind of optimization, in order to construct a system for a task.
We also include a third category classified as Overview articles, which include publications that do not involve a specific implementation, but instead present a review or analysis.

Special thanks to Brecht De Man, Joshua D. Reiss, and Ryan Stables as their publication Ten Years of Automatic Mixing served as the foundation.

1 Previously there was a distinction made between Knowledge-engineered and Grounded theory approaches, but due to the somewhat ill-defined differentiation between these approaches, and for clarity, we choose to group all of these systems under the Knowledge-based systems (KBS) category. Since both approaches fullfil the requirements for the definition of KBS, i.e. they incorporate some level of domain knowledge in conjunction with an optimization process or algorithm for the application of this knowledge, this seems to be a fair classification.