When is Noisy On-Set Dialogue Worth Replacing? AI vs. ADR

The threshold for switching to ADR is when noise reduction artifacts compromise dialogue intelligibility or emotional fidelity more than the original noise, even with the best AI tools. I've had to make this call countless times over the last decade, often with incredible performances that fight against a noisy backdrop. For instance, on a recent indie feature shot with an ALEXA 35, we had a pivotal scene where an actor gave an unbelievably raw performance, but a distant, unidentifiable rumble (turned out to be a delivery truck idling two blocks away) was consistently present in the 4017 boom track. My initial move is always to try cleaning it up, I'll often start with iZotope RX or SpectraLayers, meticulously painting out frequencies. But there's a point where aggressive noise reduction starts to thin out the voice, making it sound hollow or phasey, losing that organic texture that connects the audience to the performance. Once I hear that vocal 'chatter' or a loss of presence that distracts from the delivery, even if the words are technically clear, that's when I start pitching ADR to the director and supervising sound editor. At that point, the 'character' of the voice is suffering too much. ADR, when done well, can re-capture that performance with pristine audio, allowing the editor and mixer to truly sculpt the emotional impact. The guide, "Production Dialogue and ADR: The Definitive Workflow for Filmmakers," really emphasizes this integration, highlighting sound as a core storytelling element from beginning to end. https://blockreeldao.com/blog/production-dialogue-and-adr-the-definitive-workflow-for-filmmakers

What specific tell-tale signs do you listen for that push you over the edge into ordering ADR, especially when an actor knocked it out of the park on set?