ElevenLabs Puts Voice Agents in an F1 Wind Tunnel to Tackle Background Noise

ElevenLabs brought voice agents into the Audi F1 wind tunnel.

In the video, a person walks down the center of a circular wind tunnel, facing away from the camera, with light coming in from the far end. There are vertical stripes on the tunnel walls. ElevenLabs says to driver Hulkenberg: Sorry, it's too loud in here.

Voice agents don't usually operate in a quiet test room. When customers call, there may be a TV on nearby, a dog barking, or the roar of a racing grandstand. ElevenLabs wanted to try a more extreme scenario, so it borrowed Audi F1's wind tunnel. F1 wind tunnels are used to test aerodynamics; they have high wind speeds and plenty of noise, making them a naturally extreme environment.

Someone in the comments hit the key point. A user said background noise is currently the 'deadliest killer' for ElevenLabs agents and asked whether the relevant processing was going to be updated. Another user said customers won't wait for the TV to be turned off or the dog to stop barking before they start talking, so testing in a noisy environment is closer to reality.

Other users were curious whether, in the wind tunnel, the agent could distinguish wind from human speech and allow users to interrupt at any time. Another took it further: Could you use ElevenLabs to generate voices and synchronize lip movements with video generation to make your own movie? Another asked whether ElevenLabs could partner with F1 to remove noise from team radio communications so viewers can clearly hear the drivers and engineers.

These questions point in different directions, but they all lead to the same thing: for voice agents to be usable, they first have to withstand real-world noise. Clearly hearing commands in a quiet room is no great feat; being able to hear them in a wind tunnel is. TV sound, barking dogs, and grandstands are all microcosms of this wind tunnel.

At the very least, this test shows that ElevenLabs is taking this 'killer' seriously. As for whether background-noise processing will be updated, those follow-up questions in the comments have already asked on everyone's behalf.