SRT generator — audio, video, or a URL in, a timed .srt out

What you get

A standards-clean .srt: sequentially numbered cues, comma decimal separators in the timecodes as the format requires, and UTF-8 encoding without a BOM — so accented characters survive and no player chokes on the first cue index. Multi-speaker recordings carry speaker labels on every cue. The same processing also exports VTT for HTML5 video and plain TXT, at no extra cost.

1
00:00:00,220 --> 00:00:03,480
[SPEAKER_00] And so, my fellow Americans,
ask not what your country can do for you

2
00:00:03,480 --> 00:00:06,910
[SPEAKER_00] ask what you can do for your country.

Cue boundaries sit on real word timestamps returned by the transcription engine — not interpolated across a segment — so the text appears and clears in sync with the speech.

Cue timing and reading speed

The hard part of generating a usable SRT isn't transcription accuracy — it's cue timing. A 30-second monologue can't be one subtitle, and a two-word reaction shouldn't flash for a quarter of a second.

How the splitter decides where a cue ends

Four rules, applied in order. Cue boundaries land on the word-level timestamps the model returns rather than on interpolated guesses, so a cue starts and ends where the speech actually does — the difference between subtitles that track the audio and subtitles that drift a few frames further out with every cue.

A very short trailing cue is merged into the one before it rather than left dangling, so the occasional cue runs slightly past those caps. Speaker changes always break, which is why multi-speaker recordings come back as readable exchanges instead of one wall of text with everyone's words run together.

Why word-level timing matters here

Most tools stamp a block of speech every few seconds and cut cues on a fixed character count, wherever that lands. Splitting on word timings instead means a cue can end on the last word of a sentence rather than mid-clause, and a pause between speakers becomes a cue boundary rather than a gap inside one. It is the same word-level data that powers the JSON export and click-a-word playback in the editor.

What the professional standards say

These are industry reference points, not our settings — the figures broadcast and streaming captioning is graded against. Check the output against them when a cue looks dense, and split it in the editor before exporting.

These figures come from the Netflix Timed Text Style Guide and the BBC Subtitle Guidelines, the two most cited references for caption timing in professional production. The practical result is an SRT that imports into Premiere Pro, Final Cut, DaVinci Resolve, CapCut and YouTube without manual cue-splitting cleanup. For how the underlying timestamps behave across audio conditions, see Whisper transcription accuracy.

After the SRT

One processing run, several outputs. Translate the subtitles into another language with the cue numbers and timecodes preserved, so the translated file drops into the same timeline. Generate an AI summary of the transcript, or ask questions about it in AI chat. Export DOCX to hand to an editor, VTT for HTML5 players, JSON with word-level timestamps for a developer pipeline, or plain TXT.

For the translation workflow in depth — including uploading an SRT you already have — see subtitle translator. For timestamp formats and the word-level JSON structure, see transcription timestamps.

Audio-to-SRT tools compared honestly

There are genuinely free, genuinely unlimited SRT tools, and for a short one-off file they are a reasonable choice. The difference shows up in the cues rather than the words.

The honest read: if you need one short file captioned and you are willing to fix the timings yourself, a free unlimited tool will do it. What you usually get back is the transcript as a single long cue, or cues cut on a fixed character count with no regard for who is speaking — fine to read, tedious to sync. Running Whisper yourself gives you better text but the same chunked cues, plus a separate diarization step if you want speaker labels. Our 5-minute preview exists so you can compare the actual output rather than take any of this on trust.

Subtitling one video in several languages

Generate the source-language SRT once, then translate the cue text while the timecodes stay exactly where they are. The translated file drops into the same timeline without re-syncing, which is the part that usually costs the most time.

1. 1Generate the SRT in the spoken language.
2. 2Translate it — cue numbers and timecodes are preserved, only the dialogue text changes.
3. 3Export each target as video.<lang>.srt using BCP-47 tags (video.es.srt, video.de.srt, video.pt-BR.srt).
4. 4Reference every language track in your player, HLS manifest, or YouTube upload.

Free translations are included with an account. Premium translation is the better choice for long recordings, where it keeps terminology and proper names consistent from the first cue to the last instead of drifting partway through.

Accessibility and legal compliance

WCAG 2.1 Success Criterion 1.2.2 (Level A) requires synchronized captions for all prerecorded video audio. The U.S. DOJ ADA Title II rule sets binding deadlines: April 26, 2027 for state and local governments serving 50,000 or more people, and April 26, 2028 for smaller entities.

Beyond compliance, caption text is indexable, and a large share of social video is watched with the sound off — captions are what keep those viewers watching. See our editorial and accuracy standards for how we test.