Task

To research and design new AI-powered sound mixing tools. These tools aim to automate repetitive yet complex and somewhat subjective/artistic decisions during the sound mixing process of movie soundtrack.

Challanges

• Working with large datasets (hundreds of gigabytes) but at the same time sparse in terms of useful information
• Creating completely novel AI tools with no prior art available as a reference

How we helped

• Created and trained several neural networks models that which mimic the actions of a sound engineer (Pytorch)
• Created an automated machine learning workflow for ingesting data, training models and assessing performance (MLFlow)

Results

Our contributions helped our client, Skywalker Sound (Lucasfilm), automate several aspects of movie soundtrack mixing, saving them both time and money. More information can be found here: https://movielabs.com/case_study/skywalker-keeps-the-humanity-in-automated-soundtrack-mastering/