Add subtitles to video

A speech model writes the words it knows. Software tutorials are full of words it does not: product names, command-line tools, keyboard shortcuts and acronyms. “Kubectl” comes out as “cube control”, “GitHub” as “get hub”, “npm” as “NPM” or “and PM”. Read through the subtitles once with the video playing; that is the only way to catch these.

When the same word is wrong many times, add it under Fix a name or term everywhere: what the model wrote, and what it should say. The fix replaces whole words only, in any letter case, and finds a phrase even when it is split across a subtitle’s two lines. The list stays in your browser, and each new video’s subtitles are corrected with it as they appear. It is a find-and-replace on the result, not a way to teach the model, so check the first few subtitles of each video anyway.

SRT (SubRip) is the simplest subtitle file: a number, a start and end time, and the text. YouTube, Vimeo, LinkedIn, Premiere Pro, DaVinci Resolve and most video players read it.WebVTT (.vtt) is the format a web page’s video player needs. Both files from this page hold the same subtitles and times; if you already have one, the SRT to VTT converterturns it into the other.

Strictly, captions are the words in the language spoken, for viewers who cannot hear the sound, and should also note important sounds; subtitles are a translation. This page makes captions of the speech; add sound notes such as [music] or [door closes] by hand where they matter. Both are usually closed: viewers can turn them on or off. Captions burned into the picture areopen and always show, which suits short social clips that play without sound.

• YouTube Studio: leave automatic captions on, then open Subtitles, choose the automatic track and press Duplicate and edit to correct it.
• CapCut: Captions, then Auto captions writes captions in the editor and can burn them into the exported video.
• Premiere Pro: the Text panel’s Transcribe and Create captions make a caption track you can export as SRT or burn in.
• DaVinci Resolve Studio: Timeline, then Create Subtitles from Audio.
• By hand: the SRT subtitle editor lets you type and time subtitles against your video, or fix a file another tool made.

• Files up to 4.0 GB and 3 hours. On a phone, choose the Small model; it needs less memory.
• Speed depends on the device. On a 2020 MacBook Air (M1, 16 GB), a 2½-minute tutorial took 22 seconds with the Standard model on the graphics card, once the model was downloaded. On the processor alone, expect about half that speed.
• The first run downloads the speech model, 45 MB to 150 MB. Nothing is downloaded until you press Make subtitles.
• It makes subtitle files. To burn captions into the video, use add captions to video.
• No speaker names, no sound descriptions and no translation.
• Timing is per phrase. A long phrase is split by length of text, so check where its subtitles change.
• Mistakes are certain, more so for names, numbers, jargon and accents. Read the subtitles before you publish them.