The smell of wet wool and cedar shavings always lingers in the corners of my studio, a scent that most people would find suffocating but which I find necessary for the work. I was kneeling on a patch of distressed hardwood, grinding a handful of coarse river gravel into a shallow tin tray, trying to replicate the sound of a man walking across a Parisian terrace in 1964.

Claire T.-M. does not just record sounds; I find the marrow inside the noise. My hands were grey with silt, and the microphone was positioned exactly 4.3 inches from the tray-an idiosyncratic distance that I’ve found captures the low-end “thud” of the heel without losing the “skitter” of the loose stones.

It was in this state of granular focus that I took the call from Sophie. Sophie is my counterpart in Lyon, a woman who hears the world in much the same way I do, which is to say, we both find silence suspicious. We were supposed to be discussing the restoration of a lost New Wave reel, a project that required a surgical level of acoustic matching. Because my French is a patchwork of restaurant phrases and technical jargon, and her English is similarly fragmented, we were using one of those high-end translation suites that promises “seamless connectivity.”

01

The Turn-Taking Tax

The software sat between us like a polite, invisible butler. For the first ten minutes, it was civil. I would speak a sentence, wait for the little blue orb on the screen to pulse, and then Sophie would hear the translated version in her headset. She would respond, I would wait, and the cycle would repeat. It was orderly. It was clean. It was also, I quickly realized, the fastest way to kill the creative spark of a conversation.

We reached a point in the discussion where we were both looking at the same frame-a close-up of a glass of red wine shattering on a cobblestone street. I had an idea about layering the sound of a breaking lightbulb with a recorded snap of a dry celery stalk. Sophie had a counter-idea involving crushed seashells. We both started talking at once. I was saying, “Wait, the celery gives it that organic crunch,” and she was simultaneously shouting, “Mais non, Claire, the shells have that vitreous, sharp ring we need!”

The software, designed for the “polite” exchange of information, couldn’t handle the collision. It tried to parse both of us at once, got confused by the overlapping frequencies, and simply stopped. It produced a garbled string of text that looked like a cat had walked across a keyboard, and then it fell into a stubborn silence. By the time the system reset and signaled that it was ready for Speaker A to resume, the heat was gone.

The shells and the celery felt like distant, academic concepts rather than the urgent, visceral solutions they were ten seconds earlier. I’ve spent 18 years as a foley artist, and if there is one thing I have learned about sound, it is that reality is never sequential.

In a real room, people talk over each other. We finish each other’s sentences not out of rudeness, but out of an impatient, beautiful empathy. We use “back-channeling”-those little grunts of “uh-huh” and “yeah” and “right”-to signal that we are still on the bridge together. Most translation tools treat these as errors to be filtered out, as if the goal of communication is a clean transcript rather than a shared moment of understanding.

The Interview vs. The Conversation

I’ll admit, I was wrong for a long time about what constituted a “good” recording. I used to think that the highest form of audio was the isolated track, the voice stripped of its environment, pure and sterile. I spent the better part of a decade trying to eliminate the “bleed”-that accidental sound of one instrument leaking into another’s microphone.

But I eventually realized that the bleed is where the music actually lives. It’s the same with conversation. If you strip away the interruptions and the overlaps, you aren’t left with a better conversation; you’re left with an interview. Or worse, a deposition.

The designers of these orderly systems model conversation as a game of ping-pong. Player A hits the ball, Player B waits for the ball to land, and then Player B hits it back. But real rapport is more like a jazz session. We are playing notes at the same time, sliding into each other’s measures, creating a third sound that neither of us could have produced alone.

I recently waded through the terms and conditions of three different live-translation platforms-an exercise in masochism that I don’t recommend-and noticed a recurring theme. They all boast about “noise cancellation” and “speaker isolation.” They are obsessed with the idea that the “other” person is noise. But in a lively dialogue, the other person’s interruption is the signal.

This is why the shift toward v2.0 speech models is so critical for people who actually need to get things done. When I started looking for a tool that wouldn’t kill my workflow with Sophie, I realized that the secret isn’t just better vocabulary; it’s lower latency and the ability to handle the “mess.”

You need something that doesn’t panic when two people get excited. For instance, Transync AI is engineered with a sub-0.5-second latency-specifically around 0.38 seconds in optimal conditions-which is the threshold where the human brain stops perceiving a delay as a “wait” and starts perceiving it as a “flow.”

The Goal of Invisibility

If you can get the translation to happen in less time than it takes to blink, you can actually maintain the rhythm of a high-involvement conversation. You can interrupt. You can say “Exactly!” while the other person is still finishing their point, and the system won’t collapse into a pile of digital rubble.

“We were overlapping constantly. The translation was keeping up, feeding us the essence of each other’s words in real-time. Because the system didn’t force us to wait, we forgot the system was there.”

– Claire T.-M., on a session with a Japanese Director

I remember a specific session where I was trying to explain the sound of a “heavy” silence to a director who only spoke Japanese. We were using a more advanced, low-latency setup this time. I wasn’t waiting for the orb to pulse. I was describing the sound of a radiator hissing in an empty apartment, and he was jumping in with his own descriptions of a cold wind through a cracked window.

When software is slow, it dictates the culture of the meeting. It makes everyone “act” professional in a way that is stifling. You see it in Zoom calls where people sit with their hands folded, waiting for their three-second window to speak. It’s a performance of order that masks a total lack of creative friction. I hate that version of the world.

I’d rather have a messy, loud, overlapping argument where something new is actually created than a perfectly transcribed sequence of polite platitudes. There’s a technical term for what Sophie and I do: “cooperative overlap.” It’s common in many cultures-from the Mediterranean to the Bronx-where talking over someone is a sign of intense listening.

If you don’t interrupt, it’s assumed you’re bored. The translation tools of the last decade were built by people who seemingly value a quiet, linear Anglo-centric model of “taking turns.” They built a tool for a library, not a workshop.

Distinguishing the Collision

But the world is a workshop. It’s full of gravel and wet wool and people who have something urgent to say. We are currently seeing a move toward models that can distinguish between “competitive interruption” (where someone is trying to steal the floor) and “cooperative overlap” (where someone is building on the current speaker’s point).

That distinction is everything. It’s the difference between a tool that assists you and a tool that manages you. I went back to my gravel tray after the call with Sophie. I realized that the reason the “ping-pong” translation felt so wrong was that it removed the “texture” of our relationship. Sophie’s interruptions are part of her vocabulary.

When she cuts me off to tell me my idea about the celery is brilliant but needs more “wetness,” that interruption is a gift. It’s a shortcut. When a machine removes that shortcut, it’s adding miles to the journey.

I’ve started using the newer v2.0 models in my international sessions now. The difference is subtle but profound. It’s the difference between walking on a paved sidewalk and walking on that river gravel. On the sidewalk, you know exactly where your foot will land. It’s safe. It’s predictable. But on the gravel, there’s a bit of slide. There’s a bit of grit. You have to be more present. You have to listen to the sound of the skitter.

In a recent meeting with a studio in Berlin, we were arguing about the sound of a 19th-century carriage wheel. We were three people, all talking over each other in a mixture of German, English, and purely onomatopoeic sounds-“crrr-ack,” “whirr,” “thump.” The translation was humming along in the background, catching the nouns, ignoring the “uh-huhs,” but most importantly, not stopping the clock.

We weren’t Speaker A and Speaker B. We were just three people trying to find the sound of a wooden spoke cracking under the weight of a ghost. We found the sound in about six minutes. In the old, sequential model, that would have taken twenty. We would have lost the “whirr” while waiting for the “thump” to be translated.

A Form of Suppression

I think about the designers of these systems often. I wonder if they ever sit in a room where everyone is shouting with joy. I wonder if they realize that their “orderly” model is actually a form of suppression. When we force people to wait for the machine to finish, we are telling them that the machine’s processing time is more valuable than their human momentum.

I’m a foley artist. My entire life is built on the belief that the “mess” is the message. The rustle of a silk dress, the squeak of a floorboard, the overlap of two voices in a heated moment-these aren’t things to be cleaned up. They are the things that tell us we are alive.

If we’re going to use AI to talk to each other across the world, the least we can do is demand an AI that can handle a little bit of noise. Because without the noise, we aren’t really talking. We’re just taking turns being alone.

I picked up the iron bolt from my workbench and slid it home. Clack. It was a perfect, solitary sound. But then I imagined Sophie’s voice jumping in right before the click, telling me it needed to be heavier, more final.

I smiled. That would be the better version.