Issue link: https://digital.copcomm.com/i/1504940
C A S Q U A R T E R L Y I FA L L 2 0 2 3 13 Photo courtesy of M:NI AI is an essential post-production audio tool: analyzing production dialogue, identifying human voices, and effectively removing countless bumps, ticks, and hums from production noise. AI has become so efficient at cleaning dirty production that many tracks declared NG in the past now go on-air without extra takes or ADR. On the music side, AI is able to transform raw tracks into relatively polished and mastered cues. Knowing these capabilities exist, one wonders how soon AI might be employed to create complete final mixes. The technical requirements of balancing tracks, containing levels, and meeting specifications seem well within reach, but the biggest hurdle seems to be interfacing with the writers, directors, and producers to create the dramatic narrative moments necessary for effective audio storytelling. Steve Borne, CEO of HearLabs, audio inventor, sound designer, editor, musician, and re-recording mixer joined me on Zoom for a lively discussion about the future of audio AI and the power and perils of using it more comprehensively in post. Steve Borne: (Gesturing behind him) There's a patent behind me. I hold patents in the audio space about clarity in audio and what people hear and don't hear. So, this is a lifelong artistic creative pursuit for me. Would I be interested and happy enough to hear an AI mix? Of course I would, because that's part of my lifelong pursuit of sound. But at the same time, I think the human factor is super important. So, if there's an AI movement, there's likely to be an anti-AI TV movement at some point because people are going to realize AI ain't Shakespeare. I'm an AI fan. I like AI for a lot of things [like] playing chess [or for things such as the] IBM Watson project [where] Watson can ingest every medical text ever written in about five minutes. [If] a kid's got a funny mole in a funny spot and instead of turning red, the ring around it turns brown, [with] those three symptoms, Watson can cross-reference and give an amazing list of things for doctors to look at. Kurt Kassulke CAS: I have a guy on my block who does startup proposals. He's set up a ChatGPT with a form you fill in, and it cranks out a document that does the work of four assistants in a fraction of the time, and is 90% ready to present. So, AI is clearly good at managing tasks, but I have my doubts about it managing art. SB: In order for AI to do sound, the directors and producers would have to be able to describe what they want well enough for the AI to actually do it as well as you and I do. A lot of our work is, we play it for them, they hear it, and then they like it. KK: We'll know it when we hear it. SB: We're meant to be the trusted source, right? We're meant to be yelled at if we do something they don't like us to do. Where does that leave anybody trying to do a creative pursuit with AI? If you're using it to sort of follow people's actions, I think it's great. But there are some things with respect to sound that are kind of unlearnable, right? It happens in the moment. There's a famous story about Leland Sklar, the bass player. He's got a switch on his bass that doesn't do anything. So, when a record producer is sitting there saying, "Hey, the bass sounds a little funny," he flips the switch and says, "How's that? Okay?" So the thing about AI for sound and the reason why AI is bad for sound is that there's this effect, the training effect. You heard it once and you're in love with it, or you heard it once and you hate it. Then you play it back the second time. The exact same way. And they're like, "Yeah, that's much better." How does an AI deal with that? It could not. The AI would assume we didn't like the first one and actually make a change. KK: On my stage, we call it an "air fix," and we will fess up. We'll turn back and go, "I'm sorry, I didn't change anything." SB: I know some AI guy's going to figure this out eventually, but with respect to this kind of thing, I want active listening, right? When we have to mix something, we go in a room, we turn off the fans, we close the door, and we pay attention. There is so much that happens in real time in every moment whether it's additional noise or whether it's not being focused or it's not having a treated room—there's so much in there. I think it's a really hard task for AI to correct. ChatGPT