ReviewCoach
All posts
PRODUCT3 min read

Syncing two phones by ear

Two dancers filming the same run-through never press record at the same moment. An early prototype lines their clips up automatically, using nothing but the song both phones happened to capture.

By Haytham Labrini (@playfulhay)

Two dancers film the same run-through — same song, same room, same eight-count — and their phones never start at the same instant. One presses record a beat early. The other fumbles with the case for two seconds. The clips are of the same performance, but you can't watch them side by side until someone scrubs both timelines into alignment by ear.

We've been prototyping a way to skip that step entirely: upload both clips, and let the audio line itself up.

The song already knows when you started

Both phones captured the same music, through two different microphones, with two different amounts of dead air stacked in front of it. That shared recording is a fingerprint. Line up the audio, and the video riding along with it lines up too — a phone's own audio and video never drift apart from each other, only the two files drift from each other.

How cross-correlation lines up two phones' audioPhone A and Phone B recorded the same music, but Phone B has a longer quiet lead-in before the song starts, roughly 13.5 seconds later than Phone A. Sliding Phone B's audio until its song pattern matches Phone A's finds that offset. Once shifted, both waveforms line up exactly, and the video riding along with each phone's audio lines up too.Phone APhone B13.5s offsetslide until the songpatterns line upAligned
Same song, two phones — sliding one waveform across the other until the pattern matches finds the offset, and lines up the video riding along with it.
Animation of a waveform sliding into alignmentA simplified, looping illustration of cross-correlation: a copy of the waveform slides in from the right and searches for the position where it lands exactly on top of the reference waveform. Once it does, the label switches from "searching" to "matched" — that landing position is the offset between the two phones.searching for the offset…matched ✓
A simplified loop of the search: the moving copy slides until it lands exactly on the reference waveform — that landing spot is the offset.

Built to be honest about its confidence

We tested it against real footage — two phones, same room, thirteen and a half seconds apart — and it found the offset exactly. It also reported how sure it was, and the number was lower than a lab-clean test signal would produce. That's by design, not a flaw: real microphones pick up room noise and sit at different distances from the music, so real footage should score lower confidence than a studio-clean recording. A number that always claims certainty isn't trustworthy. A number that's honest about real audio is.

Where this is headed

This lives today as a lab prototype, separate from the groups and queues already in the app — worth proving on its own before it earns a place in the review flow. The idea is simple, though: for anyone comparing more than one take of the same routine, the moment two clips agree on when the music started is the moment you can finally watch them side by side and see who's actually behind the beat.

Give your footage somewhere to go.

Start a group free for up to 10 people — no card, no student accounts to pay for.

Start your group