Issue link: https://digital.copcomm.com/i/1546135
58 FA L L 2 0 2 6 I C A S Q U A R T E R L Y What Happens If We Separate Production Sound Into Voice, Reverb, and Noise in the Dialogue Edit—and Preserve That Separation Throughout the Mix? Last year at Sounding Board v5.0, I attended the "Surviving AI" panel, and one thought stayed with me: If and when we incorporate machine learning into our craft, it should 1. replace no humans, 2. perform only non-creative tasks, 3. not lead to less work, but better results. So, let's say we fulfill the first two conditions by hiring the same crews and automating only highly repetitive tasks. Where exactly will the better mix results come from, other than by having more time to mix creatively? Can we automate tedious tasks in ways that change the creative process that follows? Can we create a different starting point altogether—one that allows us to work in a more playful and exploratory way? I found that permanent dialogue separation could be one of the answers. Why Separation Matters At its core, mixing is built on separation. Our craft depends on the ability to isolate elements, control them independently, and combine them into a coherent whole. Arguably, the more meaningful separation we can achieve before a mix begins, the greater our creative control will be. We already go to great lengths to produce score in separate instruments and stems, Foley and sound effects in separate layers and groups, dialogue with separate microphones for individual characters. Yet some forms of separation are temporary by nature and are effectively undone before the mix. Noise reduction, for example, is a form of separating dialogue from noise. De-reverbing is a form of separating dry signal from wet. Modern machine-learning tools such as Hush Pro, Supertone Clear, iZotope RX Dialogue Isolate, Acon Digital Extract:Dialogue, Accentize dxRevive, and Cedar Audio VoicEX have dramatically improved dialogue editorial and rescued PERMANENT DIALOGUE SEPARATION: Isolating Voice, Reverb, and Noise Across All Stages of Post for Improved Control b y G r e g o r B o n s e C A S production recordings that might previously have required ADR. But have they expanded the creative possibilities of the mix itself? I would argue that they have not. The separation achieved by these tools, whether used in editorial or on the mix stage, rarely survives past the render button. Once processed through AudioSuite, the relationship between voice, reverb, and noise is effectively locked in place before mixers ever touch a fader. And for many other tasks, that is perfectly acceptable. Destructive processing remains an efficient solution for those artifacts we either want to preserve or remove entirely, such as distortion, crackles, clipping, clicks, and pops. Voice, reverb, and noise, however, are different. All three are fundamental components of the production track, and each serves a distinct purpose within the mix. They do not require removal. They require balance. If mixers wish to define this balance, they are routinely left with two options. They can render clips from the mix stage, each time interrupting the creative flow as they move back-and- forth between console and mouse. Or they can work with real- time tools, much like they had in the days of the Dolby Cat 43 and early Cedars. Since the arrival of iZotope RX, however, technology's momentum seems to have largely shifted toward AudioSuite and offline workflows. While most manufacturers offer real- time alternatives, most of them operate at lower capability or quality than their offline counterparts. Notable exceptions are Cedar VoicEX and Accentize dxRevive Pro, though they do not separate reverb. That limitation led me to explore a third possibility: an approach that separates production sound with the full potential of AudioSuite tools but maintains that separation throughout the entire sound post. As a result, mixers end up with three dedicated VCA's on the console: one for voice, one for reverb, and one for noise.