Issue link: https://digital.copcomm.com/i/1546135
60 FA L L 2 0 2 6 I C A S Q U A R T E R L Y Dialogue editorial began by sorting through the fully referenced dialogue build and transferring booms and lavalieres to their corresponding tracks. Because picture editorial had largely maintained dialogue on a single track with crossfaded edits, most referenced material initially landed in the A pack. At this stage, we deliberately avoided introducing checkerboarding. All lavalieres were phase-aligned to their corresponding booms using Auto-Align Post 2. Processing was performed clip-by-clip with three-second handles across the entire timeline. On this project, the process worked flawlessly, although earlier versions of the plugin had occasionally produced hiccups that required manual correction. The boom tracks were then separated into voice, reverb, and noise. For the separation, we used Ian Sampson's Hush Pro, which, at the time of writing, comes closer than any other tool I have tested to achieve convincing separation. Anyone interested in exploring this workflow should, of course, feel free to use other tools. The larger concept is not tied to a particular plugin—the idea is permanent separation as a creative framework. Even so, no current solution is perfect. Hush can still produce artifacts during aggressive processing, particularly on breaths, where the results occasionally take on an almost reptilian quality. Encouragingly, Ian Sampson recently told me that a Version 2 release is planned for this fall and is expected to address this very issue. Counterintuitively, Hush Split itself proved unusable for our purposes as it supports only "create continuous file" and entire duration processing due to Pro Tools' limitations of multi- channel files. To preserve fades, handles, and editorial flexibility, we built a SoundFlow macro that processed all boom tracks globally at the click of a button. Automating the Separation The macro workflow was straightforward: 1. Copy and mute A boom. 2. Paste the material onto A —voice, A —reverb, and A — noise. 3. Process Hush Mix Pro three times across the entire timeline using clip-by-clip with three- second handles—once each for voice, reverb, and noise. We chose not to separate the lavalieres. Because they were unlikely to contain reverb or production noise worth preserving as independent mix elements, they were later cleaned in a more traditional workflow. Thanks to the neural engines in Apple Silicon machines, machine-learning processes such as Hush can operate remarkably fast. Processing an entire feature still involves waiting through a considerable number of progress bars, but that was, in many ways, the point. Using a SondFlow macro, boom tracks are split into voice, reverb, and noise, while preserving three-second handles The Experiment I first tested this workflow on Eurotrash, directed by Frauke Finsterwalder and adapted from the novel by Christian Kracht. My first call was to Gaston Ibarroule, whom I wanted as dialogue editor on the project. The production track layout was going to be more extensive, immediately revealing perhaps the biggest eyebrow-raiser of the experiment: The system was built around four track packs—A, B, C, and D—each containing: • Boom • Voice • Reverb • Noise • Lav 1 • Lav 2 • Lav 3 • Lav 4 • Lav 5 • Lav 6 In total, the production track layout expanded to 40 tracks, excluding PFX and ADR.