How to Find Clip Timestamps in a Transcript (SRT/VTT)

Short answer

This guide is for footage you own or have the rights to (your podcast, webinar, tutorial or stream). It does not cover taking other people's videos.

Step 1: Get a transcript of your own video

If the video is on your YouTube channel, you can download its caption track from YouTube Studio. YouTube Help describes the steps: sign in to YouTube Studio, choose Subtitles in the left menu, click the video, then under Subtitles click Edit next to the language, open Options, and select Download subtitles. The file you get is a plain-text caption file you can open in any text editor.

Two cautions from the same help pages. First, automatic captions are produced by machine learning, and YouTube says they "might misrepresent the spoken content" because of mispronunciations, accents, dialects or background noise, and that you should always review and edit them. Second, automatic captions can be missing: YouTube lists reasons such as the video being too long, poor sound quality, overlapping speakers or unsupported languages. If you edit an automatic track, YouTube generates a new track with your revisions.

If the video is not on YouTube, any transcription tool that exports SRT or WebVTT works, and a hand-typed list of mm:ss text lines works too. The method below only needs words and times.

Quick review habit: read the transcript once for wrong names and numbers. A misheard number is exactly the kind of thing the checklist below would rank highly.

Step 2: Read SRT, WebVTT and mm:ss lines

All three formats are just text with times. Here is the same moment in each.

SubRip (.srt)

Per the SubRip description, an .srt file holds plain-text subtitles in groups separated by a blank line, numbered from 1. The timecode is hours:minutes:seconds,milliseconds, with a comma before the milliseconds.

1
00:05:12,400 --> 00:05:16,900
Why do most first loaves come out dense?

2
00:05:17,200 --> 00:05:23,800
Because people add flour until the dough feels easy, but that is the mistake.

WebVTT (.vtt)

The W3C WebVTT specification says a file begins with the string WEBVTT. A cue's timing line has a start timestamp, the string -->, and an end timestamp that must be greater than the start. A timestamp has two-digit minutes and seconds, a full stop, and three digits of milliseconds; the hours part is optional unless the hours are non-zero.

WEBVTT

05:12.400 --> 05:16.900
Why do most first loaves come out dense?

05:17.200 --> 05:23.800
Because people add flour until the dough feels easy, but that is the mistake.

Plain mm:ss lines

05:12 Why do most first loaves come out dense?
05:17 Because people add flour until the dough feels easy, but that is the mistake.

Mind the separator: comma in SRT, full stop in WebVTT. And note that an .srt has no single official character encoding, so if accented characters look broken, save the file as UTF-8 (YouTube's help page says SubRip files must be plain UTF-8). With mm:ss lines you only get a start time per line, so the end of a line is the start of the next one.

Step 3: Score candidate windows by hand

Work through the transcript and add a point for each signal below to the window (a run of roughly 20 to 60 seconds of lines) where it appears. These are rules of thumb, not measured laws, and you should adjust them to your own content.

1. Sentence boundaries. A window should start at the first word of a sentence and end on the last word of one. A clip that starts mid-sentence or stops before the point lands feels broken.
2. Question hooks. A line ending in a question mark near the start ("Why do most first loaves come out dense?") gives the viewer a reason to keep watching, provided the answer follows inside the window.
3. Numbers. Counts, durations, temperatures and steps ("three mistakes", "bake for 40 minutes") make a moment concrete. Check them against the audio first, since automatic transcripts get numbers wrong.
4. Contrast words. "But", "however", "actually", "instead" and "the mistake is" usually mark a correction or a twist. The sentence before the contrast is often the setup you need in the window.
5. Speech pace. Count words per second across the window's lines (words divided by the seconds between the first start and last end). Lines that are packed densely usually carry content; sparse ones are often hesitation or crosstalk.
6. Filler density. Count "um", "uh", "you know", "like" and "sort of" in the window. A window full of them is probably thinking out loud, not delivering a point.
7. Silence gaps longer than 3 seconds. In SRT or VTT, subtract one cue's end time from the next cue's start time. A gap over about 3 seconds is a natural break: avoid putting one in the middle of a clip, and consider starting a window right after one.

Do this on the whole transcript, then sort by score. Keep the top handful and ignore the rest. One window per idea: if two windows overlap, keep the one that ends better.

Step 4: Set the start and end times

Convert the first cue's start to your start time and the last cue's end to your end time. Then adjust by ear: pull the start a fraction earlier so the first word is not clipped, and let the end run a beat past the final word.

Cutting is a separate step from finding. With ffmpeg, a command of this shape cuts a window out of a file you own (the quoted times are the cue times from your transcript):

ffmpeg -ss 00:05:12 -to 00:05:58 -i input.mp4 -c:v libx264 -c:a aac clip_01.mp4

Then watch clip_01.mp4 with sound, as someone who has never seen the source. If you need the previous minute to follow it, start earlier.

Automating the same rules in the browser

The free Clip Moment Finder applies the checklist above for you. Paste an SRT, WebVTT or timestamped transcript, choose a clip length, and it returns ranked clip windows along with ffmpeg commands for each. It runs in your browser, and nothing you paste is uploaded.

What it does not do, so you are not misled:

• It is rule-based scoring, not AI. It counts the signals listed in Step 3.
• It cannot hear tone, laughter or emphasis, and it cannot see the video.
• It does not transcribe: you must bring the transcript.
• It does not cut or render your video; it only gives you times and commands.
• It cannot judge whether a moment is on-brand, accurate or safe to post.

So use its ranking as a starting list, then review each window yourself.

Doing this in RemixViral

RemixViral is built for the same job, end to end, on footage you own. You upload a long video (MP4, MOV, MKV or WebM, up to 120 minutes; there is no link import), an AI step finds the moments worth posting, cuts them to 9:16 with burned-in captions, and you review the clips before export. It does not translate or dub.

Every monthly plan starts with a paid trial. $5 to start: 100 credits for 3 days, enough for two source videos of about 10 minutes. Then the plan price (from $19/month) unless you cancel. Cancel anytime in Billing. Trial credits expire after 3 days. The trial covers up to 20 minutes of source video in total during the trial. Card required. Starter is $19/mo with 1,200 credits. A clip project costs 50 credits per source video up to 60 minutes, plus 1 credit per extra minute. See pricing or the comparison with OpusClip. You do not need it to use anything above.

Frequently asked questions

Sources

• YouTube Help, Edit or remove captions: https://support.google.com/youtube/answer/2734705 (read 30 October 2026)
• YouTube Help, Add subtitles & captions: https://support.google.com/youtube/answer/2734796 (read 30 October 2026)
• YouTube Help, Use automatic captioning: https://support.google.com/youtube/answer/6373554 (read 30 October 2026)
• YouTube Help, Supported subtitle and closed caption files: https://support.google.com/youtube/answer/2734698 (read 30 October 2026)
• W3C, WebVTT: The Web Video Text Tracks Format: https://www.w3.org/TR/webvtt1/ (read 30 October 2026)
• Wikipedia, SubRip: https://en.wikipedia.org/wiki/SubRip (read 30 October 2026)