Offline · nothing uploaded · no sign-up
Auto Subtitle Generator Offline
Drop your video or audio plus the transcript you already have. This studio finds the speech runs locally, matches your words to them by length, and hands you a correctly timed .srt or .vtt — in seconds, inside your browser.
1 · Your media and your words
Drop a video or audio file here
MP4 · WebM · MOV · MP3 · WAV · M4A · AAC · OGG
Limit 100 MB · decoded with your own CPU · first 15 minutes analysed
Paste or type at least five words so the cues have something to match.
Runs on this device. No upload, no queue, no account.
2 · Verify the timing
0:00 / 0:00
Violet blocks are detected speech runs, outlined boxes are your cues. Click or drag the ribbon to scrub, or focus it and use the arrow keys.
Blobs built in the browser — the file never leaves the tab.
How the timing works
Three passes, all of them inside your browser tab.
The file is read with the Web Audio API, mixed to mono and reduced to a 20 ms RMS energy envelope. The first 15 minutes are analysed in full; later cues continue the detected cadence.
A noise floor is measured from the quietest frames, then runs above it merge across the silence-cut window you choose. A music-only or silent track stops right here with a clear error instead of an empty file.
Your transcript is split into units, weighted by character length, and spread over the speech runs; boundaries that land near a real pause snap onto it. Every line survives, even when the counts disagree.
Good to know
MP4 (H.264 + AAC), WebM, MOV, MP3, WAV, M4A, AAC and OGG. MKV with AC3, HEVC-only MP4s and DRM-protected files usually fail — the tool says so plainly instead of writing a broken subtitle file.
Cue edges land within roughly a quarter second of audible pauses when speech detection is clean. Scrub the ribbon, tap a cue row to jump, and use the global offset to cancel a constant latency.
No upload, no server, no account, no client footage on a stranger's box. The .srt and .vtt files are Blobs assembled in memory and named after your source file.
FAQ
Does this transcribe my audio?
No — and on purpose. There is no speech recognition here, no model download and no upload. It takes the words you already have and gives them timecodes. If you have no transcript yet, export captions from your recorder or a caption tool first, then paste the result here.
Why are all my cues a second off?
Constant drift is usually a container or clock offset. Set the global offset to the difference — negative if the subtitles appear early — rebuild, and every cue shifts together. Random-looking drift means the transcript and the audio disagree about content.
My transcript has far more lines than detected speech runs.
Timing stays proportional and no line is dropped, but the diagnostics strip flags the mismatch and marks cues whose reading pace is unrealistic. Switching to "one cue per paragraph", removing stray lines, or raising the silence cut usually fixes it.
Will the .srt open in VLC and Premiere?
Yes — sequential cue numbers, HH:MM:SS,mmm timecodes, wrapped lines and a trailing blank line, which is what both importers expect. The .vtt variant carries the same timing with dots and a WEBVTT header for web players.
How long a file can I bring?
Up to 100 MB, with the first 15 minutes analysed in detail to keep memory sane on a laptop. Longer files still get a complete subtitle file — later cues follow the detected speech cadence, and the diagnostics strip says so.