How to create an SRT file — three routes that actually work

The format in one block

SRT (SubRip) dates to the early-2000s Windows program of the same name, which ripped subtitles from DVDs with OCR. The format it wrote was so simple it escaped the program: no formal standards body, no binary container — just numbered blocks of plain text. It is now the most widely distributed subtitle format on the internet, and every editor, player, and upload form reads it.

1
00:00:01,000 --> 00:00:04,500
Welcome to the demo.

2
00:00:05,200 --> 00:00:09,800
Each block is: a sequence number,
a timecode line, then the text.

The rules hiding inside that block: timestamps are HH:MM:SS,mmm with a comma before the milliseconds (a period makes it a VTT-style time that strict SRT parsers choke on); the arrow is exactly --> with a space on each side; blocks separate with one blank line; and numbering runs sequentially from 1. Text sits on one or two lines — two lines, ~42 characters each, is the readability standard. There is no header line (that is VTT) and no metadata of any kind.

Method 1: write it by hand (few cues, or learning the format)

For a handful of cues, a text editor is the whole toolchain. The steps are identical on every OS; only the save dialog differs:

The honest cost: hand-writing a 60-minute video takes 2–4 hours — you are listening, timestamping, and typing in one pass. Past five minutes of video, the next two methods are dramatically faster.

Method 2: time it on a waveform (the craft route)

Subtitle Edit (Windows, open source) and Aegisub (cross-platform) import the video, show its audio waveform, and let you set cue boundaries visually or by keyboard: start a cue where the line begins, end it at the pause, type the text, move on. Both write UTF-8 SubRip by default and display a live reading-speed check per cue. For a 60-minute video, expect 30–60 minutes of work — the accuracy of human timing at a fraction of the manual cost, and still the route of choice for translation and short-form work.

Whichever tool times the cues, the readability specs are the same: keep captions on screen 1–6 seconds, hold reading speed around 15–20 characters per second, break lines at natural clause boundaries, and never split a name or key phrase across cues. Conversational footage with cross-talk is where timing quality actually shows — tighten cue breaks around interruptions and late entrances, because viewers use captions to track who said what.

Method 3: generate the SRT from the audio (the default route)

For anything longer than a minute, generating the first draft from the audio is the practical move: a speech engine transcribes and timestamps in the same pass, you correct words and timing in an editor, then export SRT. Free web generators differ wildly in what “free” covers — the 2026 free subtitle generator comparison ranks every free tier by its fine print. Check it before uploading client footage:

Two things worth doing before and after generation: preload your custom vocabulary (product names, guest names, acronyms — tools like Transcrever accept a term list that guides recognition, and AI otherwise mishears exactly the words that matter most), and edit before export, because fixing a wrong name in the transcript is one edit while catching it after the file reaches an editor or upload form is a re-export.

The privacy-first variant runs the same pipeline locally: a Whisper-class engine on your own machine transcribes the file and exports SRT with nothing leaving the disk — the local video transcription guide covers the toolchain, and video-to-text converter pricing shows what the cloud equivalents meter.

When your SRT file is not working: the seven-trap checklist

An SRT that looks fine in a text editor can still fail in a player. Work down this list — the first three account for most broken files:

If timing is off by a constant amount everywhere, the file is valid — the export carried a global offset. Shift the whole file once (Subtitle Edit’s synchronization tool) instead of re-timing cues; drift that grows over the video means the file came from a different edit of the footage and needs a re-export from the final cut. For the format-level comparison behind these traps — and the one-line ffmpeg conversions — see SRT vs VTT.

Platform limits that shape the file you make

Practical read: format every file to the ~42-character, two-line, 1–6-second spec and it passes everywhere. YouTube is the destination with the most fine print worth knowing — its timestamp and chapter rules and the three ways to get a transcript from a YouTube video are covered separately.

Can I turn a plain-text transcript into an SRT?

Not by renaming, not by reformatting — a .txt changed to .srt is rejected by every player, because SRT requires timing data plain text does not contain. Your options depend on what you still have:

Have the audio? Re-transcribe with a tool that keeps timestamps and export SRT directly — the transcript itself gives the engine a vocabulary head start, so accuracy improves on the second pass.

Only the text? Force-align it: WhisperX or aeneas (both free, command-line) compute timestamps for an existing transcript against the source audio and emit a timed file. If the text carries per-line timestamps in a parseable format, Subtitle Edit can import and rebuild valid SRT structure. What no tool can do is recover timing that was never captured from nothing.

One pass, every format, nothing uploaded

VocalFuse runs a Whisper-class speech engine locally on your Windows machine: drop in any audio or video file and get a timestamped, punctuated transcript with speaker labels — then export SRT, VTT, or TXT from the same pass, so format choice becomes a download-menu decision. Nothing uploads: the engine runs on your PC, which is also why there is no per-minute meter (free tier, Basic $5/mo dictation, Pro $10/mo adds AI notes and summaries).

Because timing comes from the audio itself, the exported SRT arrives with subtitle-grade cue timing and real punctuation — the two things auto-captions most often get wrong — with the vocabulary already right on names the cloud engines mishear.

Related reading

How to create an SRT file — FAQ