Google Files Patent for Spotting Cropped, Zoomed Video Copies by Their Soundtrack
Zoom in, crop the edges, stamp a logo on top: the picture changes, but the soundtrack usually doesn't. Google's patent application uses that gap to catch altered copies of videos.
What Google's audio-first video matching does
Every time someone reposts a popular clip with the frame zoomed in, a logo stamped across it, or the edges cropped off, the picture stops looking like the original. The soundtrack, though, usually stays exactly the same.
Google's patent application describes a way to use that. The system first checks whether two videos share the same audio. If they do, it compares only the pictures around those moments, and it lowers its bar for what counts as similar, since the audio has already made part of the case.
Then it looks for a telltale pattern: do the best visual matches line up in step, moment by moment, the way two copies of the same video would? If so, it calls the videos a match. The filing describes this as a way to check an uploaded video for unauthorized use of someone else's content.
performing an audio matching between a first video content item and a second video content item to identify a matching audio segment …
Translation: The system first compares the soundtracks of two different videos to find where they overlap.
How the diagonal line gives a copy away
The method starts with sound. The system compares the audio of a video being checked (the filing calls it a query video) against a reference video from a database, and finds the stretch where the two soundtracks match.
Next it builds a heatmap, a grid with one video's timeline along each edge. Each square holds a score for how alike two moments look, based on visual fingerprints (compact numerical summaries of what a frame looks like). When two videos truly match, the high scores tend to form a diagonal line, because both play forward at the same pace. Crops, zooms and overlays smudge that line until a standard check can't see it.
So the patent narrows the search and relaxes the test:
- Look only at the part of the grid covered by the audio match, plus a margin of extra time on each side in case the clips are slightly out of sync.
- Keep only the top slice of scores (the description's example is the top 5%).
- Check whether those survivors bunch up along a diagonal. If enough do, declare a match.
The description says this would run when a standard visual check comes up empty. It can also handle audio that matches in several separate pieces, and it can report a confidence score, the type of edit it detected, and where the clips line up.
Why reposts get harder to hide on video sites
If you watch or post short videos, this is about the tug of war over reposts. Slapping a sticker on someone else's clip, or zooming in so it looks different, is an easy way to slip past an automatic check. The filing says the same matching could also help with related jobs: crediting original creators, routing revenue to the right owner, and sorting out duplicate clips.
There is a tradeoff built in. Once the audio matches, the system deliberately lowers its bar for visual similarity, which catches more copies and puts a lot of weight on the audio step. The filing describes the approach as a high-recall add-on to existing matching methods, so it would sit alongside the checks platforms already run.
Google's 835th filing in our Google coverage since May adds to a pattern of camera and media tools, joining work on livestream highlight clipping and uncertain camera focus.
Claim 1 is a tight recipe. It needs five steps: match the audio, score the pictures around that match, keep only the top slice of scores, check whether those bunch up along a diagonal, and call a match from that. A system that made its final call a different way would not follow that exact recipe.
Claim 9 is the wide one. It asks only for an audio match, a visual search area set by that match, and a visual comparison inside that area. Read plainly, it covers any system that lets the soundtrack decide where to look in the picture, with no mention of diagonals, percentages or cropping.
The diagonal-and-top-percent method is one specific way to do the job, and the broader claim reaches the whole idea of audio-guided visual checking. Whether the wide claim holds up is a question for the patent examiners.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Google patent news breaks.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
7 drawing sheets from US 2026/0313192 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →
Be the first to weigh in