| Summary: | The field of Automatic Mixing has been ripe with research over the past five years. There are now systems that emulate level and panning decisions with some sophistication, and these have been evaluated as being close in quality to o↵erings produced by a professional sound engineer. However, a real-world mix is an integrated e↵ort between several axes, and we are missing a ground-truth on how expert mixers holistically perform their art and craft. The purpose of this thesis is to fill that knowledge gap in what mixing best practices are. An exhaustive number of approaches were made to complete this endeavor, starting with a review of academic, technical and non-technical literature. A significant number of interviews with expert professional mixers followed, which led to a quantitative questionnaire in best practices. Simultaneously, over 20 di↵erent subjective evaluation tests were performed with the collaboration of medium-skilled to expert listeners. Another validation approach was through the devising of algorithms that could extract significative content from full-mixes. The pursuit of these goals involved the creation of two extensive datasets, one of multi-track unmixed material, and a second one of commercially very successful songs to be analyzed for production patterns. All these concurrent e↵orts led to the creation of a shifting collection of assumptions that crystalized into its final form as the 88-assumption database that is the kernel of this thesis. Unlike previous e↵orts, it focuses on top-quality, o✏ine, studio mixing of music. For validation process whenever possible, and summarized e↵orts done so far in terms of implementation. As a part of the testing and validation process, many implementation algorithms were developed, some of them novel, and are described together with the assumptions that inspired them. The work was organized so that it may contribute as a self-contained blueprint to many of the upcoming approaches to computer-assisted mixing. Despite the considerable length of our e↵ort, we acknowledge that there is so much more to unravel on this topic, that the present work serves only as a modest foundation stone, probably opening up more questions for further research than the ones it now closes.
|
|---|