Testing noise cancellation voice+vision AI [video]

Purpose:

Verify that Krisp’s noise-cancellation layer doesn’t break our existing VAD / turn-detection when the mic is surrounded by steady, broadband noise.

Test setup:

Open garage; dirt bike idling next to the phone to keep ~90 dB background.

iPhone 16 Pro running the current Sen TestFlight build.

No external mic; Wi-Fi slightly constrained due to distance from router.

What to watch in the clip:

-Whether speech remains intelligible while the engine’s running.

-Quick async function call with google search grounding.

-Sync between the frame overlay and the spoken prompt/response.

How Sen Works? Voice+Vision input sent at 1fps to LLM. We use daily webrtc transport and pipecat for media handling. fly.io deployment and pipecat cloud with krisp noise integration.

Why this matters? Sen is a camera-first assistant aimed at trades and field work; the mic is rarely in a quiet office. Robust noise handling is table stakes before we widen the beta.

Try it / reproduce iOS TestFlight link is available at our website www.withsen.com - Feedback on failures or weird edge cases is welcome.