Finding Short Clips in a Long Video Using Its Transcript

Why searching text beats scrubbing a timeline

Finding a good 40-second moment in a 90-minute recording by watching is slow, and fatigue sets in long before the end. Most editors end up clipping from the first half of a recording simply because that is the part they watched most carefully.

Text changes the economics. You can read a 90-minute transcript in a fraction of the running time, search for words across the whole recording at once, and compare candidates side by side. The timestamps then take you straight to the moment in your editor. You still need to watch every candidate before cutting, but you watch five or ten moments instead of the whole thing.

What makes a moment work as a clip

Before searching, be clear about what you are searching for. A clip-worthy passage usually has most of these properties:

• It makes sense without the rest of the video. No "like I said earlier" or unexplained names.
• The first sentence is a hook: a claim, a question, a surprising number or a strong opinion.
• It pays off quickly, typically within 15 to 60 seconds for vertical short-form platforms.
• It ends cleanly, on the end of a thought rather than mid-sentence.
• The visuals are usable: the speaker is on camera, nobody is reaching for a drink, the framing can be cropped.

A transcript can check the first four. Only watching can check the fifth.

Search patterns that surface hooks

Start with the topic words your audience cares about, then search for speech patterns that tend to precede quotable moments:

• Contrast and disagreement: "actually", "the opposite", "I disagree", "most people think".
• Lists and rules: "the biggest mistake", "three things", "never", "always".
• Story openers: "I remember", "the first time", "what happened was".
• Admissions: "honestly", "I was wrong", "nobody tells you".
• Numbers and money: search for digits and words like "percent", "dollars", "hours".
• Questions from a host or audience, which make natural clip openings.

Reading also catches what search misses. Skim for short paragraphs that read well on their own. If a passage makes you want to quote it, it probably works as a clip.

Keep a clip log while you search

Write candidates down as you find them, in a sheet or a plain document. A consistent set of columns makes the editing session fast and lets someone else cut the clips:

Start

The timestamp of the segment where the hook begins

End

The timestamp where the payoff sentence finishes

Hook

The first line, copied from the transcript

Payoff

One line on what the viewer gets by the end

Context check

Anything the viewer needs to know that the clip doesn't say

Status

Candidate, watched, approved, cut, posted

Copy the hook line exactly from the transcript. It doubles as a working title and as a check on whether the opening really is strong when read cold.

Worked example: five clips from a 90-minute podcast video

1. Transcribe the episode file and download the timestamped transcript and the SRT.
2. Search for the guest's company name, "mistake", "honestly" and digits. Skim the rest at reading speed.
3. Log twelve candidates with start, end and hook line.
4. Read each hook cold. Drop four that start with "and" or "so, like I said".
5. Watch the remaining eight in the editor at their timestamps. Drop three where the guest looks away or the hosts talk over each other.
6. Cut the five clips with a little padding and reframe them for vertical.
7. Subtitle each clip using the SRT method described below, then restyle the captions to suit vertical viewing.

The search and logging take far less time than watching the episode, and the editor sees moments from the last half hour that they might otherwise have skipped.

Cutting the clip cleanly

Transcript timestamps mark where recognized speech segments start and end, and they are close but not frame-exact. Treat them as directions to the right neighborhood, then trim by ear and eye in your editor.

• Start a few frames before the first word, so the clip doesn't open mid-breath.
• End after the payoff sentence has fully landed, not on the last syllable.
• Check the audio levels match across clips if you are posting a series.
• Add your editor's markers at each logged start time before you begin, so you can jump between candidates.
• Remove a mid-clip tangent with a jump cut only if the meaning is unchanged.

If your editor struggles with a very long file, splitting it first can help; see how to split a video file.

Subtitling clips with the SRT you already have

Short clips are usually watched with the sound off, so captions matter. The SRT from the full video covers every clip, but its timestamps are relative to the start of the long video, not the clip. There are three practical ways to deal with that:

1. Import the full SRT onto the long video's timeline before cutting. Many editors keep caption cues attached to the timeline, so when you cut a range out, the cues for that range come with it. Check how your editor handles this, since behavior varies.
2. Copy the cues for the clip's time range into a new SRT and shift every timestamp back by the clip's start time. Subtitle editors usually have a "shift times" or offset feature for this; editing an SRT file covers the mechanics.
3. Export the finished clip and transcribe it on its own to get an SRT that starts at zero. This is the simplest option for a handful of clips.

Short-form captions are often styled differently from standard subtitles: larger, centered, a few words at a time. Each cue in a mydubly SRT is one recognized speech segment on a single line, so for that style you will split cues and restyle them in your editor. If cues seem slightly early or late after cutting, subtitle timestamp alignment explains why and how to fix it.

Pitfalls and limits of transcript-based clipping

• Transcripts capture words, not moments. A big laugh, a reaction shot or a visual demonstration may be the best clip in the video, and nothing in the text flags it.
• Crosstalk is where recognition struggles most, and lively conversations are often where clips hide.
• A quote that reads well can play badly if the delivery is flat. Always watch before cutting.
• Clipping can strip context from a guest's words. If a clip could be read as saying something the guest did not mean, leave it or check with them first.
• Rights matter: clipping your own recordings is one thing; clipping someone else's video is a copyright question to settle before posting.

How mydubly helps with clip hunting

mydubly produces the searchable transcript and the subtitle files. Upload the long video in the browser, in MP4, MOV, WebM, MKV or M4V and up to 2 hours per file, and you receive a plain transcript, a timestamped transcript, and SRT and VTT files from the same job. The video file stays on your device; only compressed audio is sent for transcription. A 90-minute episode costs 90 credits (9¢), at 1 credit per minute.

For method 3 above, each clip is its own file with a 5-credit minimum, so a 45-second clip costs 5 credits (0.5¢). mydubly does not pick clips, cut video, reframe for vertical or burn captions into the picture; it has no speaker labels, so note who is talking in your clip log. Start with video to text, or the subtitle generator if captions are your main goal.

If the clips need to reach audiences in other languages, the short-form video translation use case covers translated captions and voices for clips. If the same transcript should also feed written posts, see turning a transcript into social media posts.

Clips meant for audiences in other languages can be dubbed separately; repurposing video into other languages explains how one recording becomes several language versions.

Clip your next long recording from the text

Take your most recent long recording, transcribe it, and build a clip log before you open the editor. Aim for a dozen candidates and cut the best five. After a few rounds you will know which search terms find the moments your audience responds to.