Transcript Segmentation and Timestamps: How They Work

What a transcript segment actually is

A transcript segment is a bounded unit of text with a start time, an end time, and content. In an SRT or VTT file, each segment is a self-contained block. The segment controls exactly when text appears on screen and when it disappears. Nothing about the segment is inherent to the meaning of the words — it is purely a display unit.

AI transcription tools generate segments automatically by detecting speech pauses, sentence boundaries, and breath patterns in the audio. The result is usually functional but not always readable: fast speakers produce dense segments that run close to the character limit, while slow speakers produce short segments with lots of white space. Human reviewing and adjustment is often needed for broadcast or accessibility-compliant use.

Timestamp format in SRT files

Every SRT segment follows this structure:

1
00:00:04,200 --> 00:00:07,800
First caption text here.

2
00:00:08,000 --> 00:00:12,500
Second caption text here.
Can span two lines.

The timestamp format is HH:MM:SS,mmm where mmm is milliseconds. The separator between seconds and milliseconds is a comma. The separator between start and end timestamps is --> with a space on each side. Missing spaces or the wrong decimal separator will cause the file to fail validation on most platforms.

Timestamp format in VTT files

VTT follows a nearly identical structure with two differences: a WEBVTT header on line one, and a period (not a comma) as the millisecond separator.

WEBVTT

00:00:04.200 --> 00:00:07.800
First caption text here.

00:00:08.000 --> 00:00:12.500
Second caption text here.

VTT does not require sequence numbers (though they are allowed). The WEBVTT header must appear on the very first line with no leading whitespace or BOM character. VTT also supports optional cue settings after the timestamp: positioning, line number, and alignment, though most platforms that accept VTT ignore these.

Why segment boundaries affect search and editing

When a video platform indexes caption text for search, it uses segment boundaries to understand context. A search for a specific term returns results with timestamp links to that exact segment. If segments are poorly bounded — one segment spanning a topic transition, for example — the search result points to a less precise location in the video.

For transcript editing, segment boundaries determine where corrections are isolated. Correcting the text in one segment does not affect adjacent segments. This makes editing efficient: you can fix a specific error without re-timing the surrounding context. The transcript editing guide covers segment-level correction in the context of a full review workflow.

Segment length guidelines

Timestamp offset calculations when reassembling segment transcripts

When you transcribe a long video in segments and then combine the SRT files, each segment file has timestamps starting from 00:00:00,000. Before combining, you need to offset every timestamp in segment 2 by the duration of segment 1, and every timestamp in segment 3 by the combined duration of segments 1 and 2, and so on.

Example with three segments: segment 1 ends at 00:28:44,312. Segment 2 ends at 00:31:22,000 in its own timeline, but in the combined file it should end at 01:00:06,312 (the 28:44 offset added). Every timestamp in segment 2 gets 00:28:44 added to it. This operation can be done manually in a text editor with careful find-and-replace, or automatically with subtitle editing tools like Subtitle Edit (free, Windows) which has a batch offset function. For large multi-segment files, the automated tool is faster and less prone to arithmetic errors.

Sequence numbers also need to be renumbered continuously across the combined file. If segment 1 ends on block 214 and segment 2 starts from block 1, the combined file should start segment 2's blocks from 215. Most subtitle editors resequence automatically on save; plain text editing requires manual renumbering unless you use a script.

Caption sync problems: causes and fixes

Out-of-sync captions are one of the most common issues when uploading SRT files. There are three distinct causes, each with a different fix.

Cause 1: the SRT was generated from a different cut of the video.If you trimmed the video after transcribing it, the timestamps in the SRT no longer match the edited version. Fix: regenerate the SRT from the edited video. Cause 2: the video player adds a display buffer. Some players have a 100–200ms display delay that makes captions appear slightly after the speech. Compensate by subtracting that offset from all timestamps using a subtitle tool. Cause 3: file encoding issues. This is more common with VTT files on platforms that expect SRT, and with SRT files that contain periods instead of commas in timestamps. Verify the file format matches what the platform expects before uploading.

The quickest diagnostic: check whether the first caption block appears at the right time relative to the first spoken word. If the offset at the start equals the offset throughout, it's a uniform timestamp shift — easy to fix with a bulk offset tool. If it drifts (correct at the start, wrong by the end), the video frame rate was different from what the transcription tool assumed — more involved, needs a subtitle sync tool like subsync.

The single most common cause of caption upload failures in practice is not frame rate mismatch — it is the period-versus-comma separator issue combined with a file renamed from .vtt to .srt without converting the content. The platform accepts the file, displays no error, and then shows either no captions or garbled ones. I've watched this happen more times than I can count. A two-second check of the first timestamp line before uploading catches this every time.