Auto Subtitle Generator Offline: Practical Tips and Common Mistakes
Reviewed by the OnlineFree.app team · Updated
Key points
- Auto Subtitle Generator Offline aligns a transcript you already own; it never converts speech into text.
- All decoding and speech detection run in the browser, so media files are never uploaded to a server.
- The default 350 ms gap and 42-character wrap suit most narration, and both are adjustable.
- Fix constant sync errors with the offset control, but investigate progressive drift in the source files.
- Media must be browser-decodable and 100 MB or smaller, so export your subtitle file before closing the tab.
What Auto Subtitle Generator Offline actually does
Auto Subtitle Generator Offline is a browser-only subtitle timing studio: you drop in a video or audio file together with the transcript you already have, it detects the speech segments locally in your browser, and it exports a downloadable .srt or .vtt whose cues line up with what is actually spoken. It deliberately does not do speech-to-text, and no media file ever leaves your device.
The layout reflects what the job really is. Media drop and script controls sit on the left, the cue table, timeline and waveform sit on the right, so matching and verifying alignment happens in one view instead of two. Violet speech blocks in the waveform ribbon show where the detector heard voices, and cue spans are drawn over them, so a glance tells you whether a line starts before or after the speaker. The .vtt output follows the format described in the W3C WebVTT specification.
You choose how the script is cut into cues — one cue per line, per sentence, or per paragraph — and the tool splits your text before aligning it to those detected speech blocks. The defaults are a 350 ms silence gap for cutting speech segments and a 42-character wrap width, both editable in the same panel as the global offset control. Everything happens inside Auto Subtitle Generator Offline, so a client footage clip stays on your machine.
Prepare the transcript before you paste it
Most bad timing jobs are really bad source-text jobs. The aligner can only place the words you give it, so if the transcript still contains speaker labels, stage directions like [laughter], or the timestamps copied out of a YouTube caption export, those characters get counted as speech and can push a cue off the beat.
If you plan to use per-line mode, decide the break points yourself before pasting: one sentence or clause per line behaves better than a wall of text. In sentence mode the punctuation does the splitting, so make sure commas and full stops sit where a subtitle should actually end — a run-on paragraph with two commas becomes one long, unreadable cue.
Watch for the under-five-words warning. A very short paste usually means the transcript belongs to a different clip. The diagnostics strip reports how many lines were aligned versus left unaligned, which is the fastest way to catch a mismatch before you spend time nudging offsets.
If you batch-record several videos in one session, planning those blocks ahead with the Free Online Schedule Maker Generator keeps transcripts and media files paired by session date, which is exactly the detail that causes mix-ups later.
How do you fix subtitles that drift out of sync?
A constant lateness or earliness is the easy case: change offset_ms. The field accepts -5000 to 5000 milliseconds and shifts every cue by the same amount, so a track that runs uniformly 600 ms ahead of the picture is corrected with a single value.
Drift is different. If the first minute lines up but the last minute does not, no constant offset can rescue it, because the cause is at the source: a transcript from a different take, an edit that trimmed a few seconds mid-recording, or a frame-rate mismatch in an exported video. Check which file the words actually came from before nudging anything.
Our working habit is to fix the offset first, then read the cue table top to bottom using the clickable timecodes to seek, watching for any cue that lands in silence or starts more than a beat after the speaker. Two or three spot checks near the start, middle and end catch almost every alignment problem.
Line length, gaps, and reading speed
The two numbers that shape readability most are max_chars_per_line, default 42, and min_gap_ms, default 350. The first controls how text wraps inside each cue; the second decides what counts as a break between speech segments when the detector slices the audio.
Set min_gap_ms too high and separate utterances merge into one long cue, because short pauses get ignored. Set it too low and a single breath splits one sentence into two cues, which reads as a stutter. For continuous narration, values around 300–400 ms behave well; for interview footage with long pauses, dropping lower keeps an answer from being joined to the question.
The diagnostics strip reports a words-per-minute reading pace warning. As a sanity check on your own settings, a 42-character cue held on screen for two seconds works out to roughly 21 characters per second, which is fast going for most viewers — so if the strip flags your pace, shorten the lines or give cues more room. Captions on prerecorded media are also an accessibility requirement under WCAG 2.2, which is another reason to favour comfortable reading speed over filling every frame.
Common mistakes we see in real projects
The most frequent mistake is pairing the wrong transcript with the media — an earlier draft of the script, or a re-recorded take. Everything looks plausible until you scrub the waveform and see cues sitting over silence.
Second is ignoring the diagnostics. A large unaligned line count usually means the media contains more or fewer spoken segments than the script has lines, so the text and audio do not belong together. Fix that before touching gap or wrap settings.
Third is over-tuning. Chasing a perfect wrap width on a 100 MB file wastes time; get the offset right, spot-check a few cues, export, and verify in your player. The .srt and .vtt buttons generate downloads named after the source file, so a re-export after a small tweak costs seconds rather than minutes.
When this tool is the wrong choice
If you do not have a transcript, Auto Subtitle Generator Offline cannot help, because it has no speech-to-text engine and no amount of adjustment will produce words from audio. Transcribe first, or export the platform's auto-captions, then bring that text back for timing.
It also does not burn subtitles into video, translate them, or style them beyond the wrap width. Files must be browser-decodable and 100 MB or smaller: mp3, wav, m4a, mp4 and webm generally work, while an MKV with AC3 audio may be rejected because the browser cannot decode it. Protected or DRM files fail the same way.
Because everything runs locally, there is no cloud copy to fall back on — close the tab and the alignment is gone. Export as soon as you are happy, and re-verify important timings in your editor or player before publishing. If you need other browser-based utilities for a project, they live alongside this one at OnlineFree.app.
Frequently asked questions
Does Auto Subtitle Generator Offline transcribe speech to text?
No. Auto Subtitle Generator Offline only detects where speech occurs and aligns a transcript you provide; it has no speech-to-text engine. If you have no transcript, run a transcription tool first, or export the auto-captions from wherever the video was hosted, then paste that text here to get correctly timed .srt or .vtt cues.
Is my video or audio uploaded anywhere when I use this tool?
No. The file is decoded and analysed inside your browser, so nothing is uploaded and no server-side copy exists. That is why it suits client footage or unreleased course material. The trade-off is that closing the tab loses the session, so export your .srt or .vtt file promptly.
What file types and sizes does Auto Subtitle Generator Offline accept?
Media must be up to 100 MB and decodable by your browser: mp3, wav, m4a, mp4 and webm usually work. MKV files with AC3 audio, DRM-protected downloads and unusual codecs often fail because the browser cannot decode them. If your file is rejected, convert it to mp4 with AAC or to mp3 first.
How do I fix subtitles that come out early or late?
Use the offset_ms control, which shifts every cue by the same amount and accepts values from -5000 to 5000 milliseconds. A uniform error — every line early or late by the same amount — is fixed with one value. If the timing drifts progressively down the file, the transcript probably belongs to a different take or an edited version.
What are the best settings for gap and line length?
Start with the defaults: 350 ms minimum gap and 42 characters per line. Lower the gap toward 250–300 ms for interview footage with long pauses, and raise it if separate utterances are merging together. Keep lines near 42 characters for a 16:9 frame and check the words-per-minute warning in the diagnostics strip.