How to Compare Only Part of a Long Recording or Video

Most comparisons start at the beginning of both files and work forwards. That is the right default — and the wrong one when the part you care about is in the middle. The edit you want to check is at 1:40. The re-recorded line is in the second verse. The scene that was re-graded starts two minutes in.

Why starting at 0:00 is not always enough

Every comparison reads a window of the file, not an unlimited amount. On DiffALL, audio is compared two minutes at a time on every plan, and a free video comparison scores 30 seconds. A recording can be up to three minutes long, so on a long file the window has to start somewhere — and if it starts at 0:00, the second half is never read.

Choosing where the window starts

When a file you choose is longer than the window, a Compare from box appears under the upload area. Type a time — 1:30, or just 90 — and the comparison reads one window from that point in both files.

The hint beside the box tells you what you will get before you submit:

• “This recording is 2:40 long and 2:00 gets analysed — start by 0:40 for the full 2:00.”
• Start later than that and it says how much is left: “From 1:10 you get the last 1:30.” That is a valid comparison, just a shorter one.
• Start past the end, or so close to it that less than a second remains, and it says so — less than a second gives a percentage with nothing behind it to check, so it is refused rather than scored.

Reading the result

Every timestamp in the result — the per-second chart, the worst moment — counts from the start of the window. So the note above the score states the range in your file’s own time: “Compared 1:30–3:00 of these files”. A “worst moment at 0:04” on that comparison is at 1:34 in your recording — add the start time to any timestamp in the result to find it in the file.

The same start point applies to both files. If the part you want sits at different times in each — say one version has a longer intro — trim one of them first so the two line up, or rely on automatic alignment if the difference is within two seconds (see comparing recordings that start at different times).

Why a short clip against a long file scores low

A related mistake comes up often enough to deserve its own section. You have a 30-second clip and the three-minute recording it was cut from, and the comparison scores around 1%.

That is not the tool failing to recognise the clip. Both files are compared from their beginnings (lined up within two seconds), so the first seconds of the clip are being measured against the first seconds of the long file — which is a different part of the recording. If the clip came from 1:20 onwards, nothing in it lines up with 0:00.

When the two files are very different lengths and the score is low, the result says exactly this. The fix is the same as above: choose the start point in the long file so the window begins where the clip was cut from, or trim the long file to the same part.

Through the API

The API takes the same start point as start_s, in seconds. Because a script has no hint box to read, the response says when a window was applied: it carries segment_start_s whenever a start was used, and analysed_seconds alongside source_duration_s whenever less than the whole file was read. A response without those fields read the whole file.