If you use MacWhisper to make subtitles, suubbs is the Mac app built for that job alone: Whisper on your own Mac, then every word timed to about 3 ms, lines broken by subtitle reading rules, a styling studio with word-by-word highlighting, and SRT, WebVTT or styled ASS out — for $9.99 once, with a free browser version to try first. If what you want is the transcript itself, MacWhisper is the tool built for that, and this page will say so.
Both are Mac apps built on Whisper; suubbs never uploads your audio, and MacWhisper's own site describes how it handles yours. What they do afterwards is different.
MacWhisper is a transcription app. It is mature, it has years of refinement behind it, and its job is the text: interviews, notes, meetings, a podcast archive. If that is what you want, it is a strong choice and this page is not going to argue otherwise. Its own site is the place to check its current features and plans.
suubbs is a subtitle studio. It cares about where each word lands to the millisecond, how long a line stays up and whether it can be read in that time, how the caption looks on the picture, and whether the styled export renders the same when ffmpeg burns it in. If what you want is captions on a video, those are a different set of problems — and that includes captions made from a podcast or interview recording, which audio to SRT walks through.
Where we have not verified a MacWhisper detail, the cell says so rather than guessing.
| MacWhisper | suubbs | |
|---|---|---|
| Runs Whisper locally | As it describes itself — check its site | Yes — whisper.cpp on your Mac |
| Built for | Transcription | Subtitles and captions |
| Price | Free and paid versions — check its site | $9.99 once; free browser trial |
| Word timing | — | About 3 ms, aligned then snapped to the waveform |
| Line breaking | — | By reading rules: 42 characters × 2 lines, or 22 × 3 for vertical |
| Caption styling | — | Font, colour, box, outline, position, five presets |
| Word-by-word highlight | — | Colour, box or grow, from measured timing |
| Styled export (.ass) | — | Yes, matching the preview |
| Languages | Whisper's range — check its site | Sixteen |
| Batch transcription | Check its site | One file at a time |
| Plain-text transcript export | Check its site | No — timed subtitle formats only |
| Free browser version | — | Yes, first 60 seconds, no account |
This is where the two tools part ways, and it is all on the suubbs side of the table.
A transcript is fine if a word's timestamp is a tenth of a second out. A word-by-word caption is not: a highlight that lands 100 ms late reads as out of sync. suubbs aligns every word with whisper.cpp's dynamic time warping and then snaps each boundary onto the nearest genuine pause in the waveform. Measured against known onsets, alignment alone was 143 ms out on average; with the waveform step it was 3 ms. The word-level timestamps page explains the measurement.
A subtitle line has a width and a reading speed. suubbs groups words into cues by those rules rather than by sentence: up to 42 characters a line and two lines a cue by default, or 22 characters and three lines for vertical video, with lines that read faster than the limit flagged in the cue list. The reasoning behind the numbers is on subtitle reading speed.
Subtitles are seen, not just read. suubbs has a style panel with five starting presets — Classic, Boxed, Pop, Highlight and Film — plus font, size, colour, a box, outline or shadow, position, and a highlight that follows the spoken word by colour, box or growing it. The preview is the video itself, with safe-area guides for TikTok, Reels and Shorts.
The export menu has SubRip (styled or unstyled), WebVTT, Advanced SubStation (.ass) and word-level JSON. The ASS file carries the whole style, word highlight included, and renders the same wherever libass renders it — which is how you get styled captions onto a video with one ffmpeg command, covered in burning subtitles into a video.
What suubbs runs, from its own source.
suubbs for Mac runs Whisper through whisper.cpp. The recommended default is Large v3 Turbo (quantised); Base and Small English-only models are there for speed. Around the model it adds Silero voice detection, so stretches with no speech in them are not transcribed, and a Names and jargon field that biases the transcript toward the spellings you give it. Spoken language can be detected automatically or chosen from sixteen. The models download once, and from then on transcription works with the network off; offline subtitle generation lists everything the app does fetch, and why.
If you are moving from any Whisper app, the accuracy you are used to should broadly carry over, because the model is the same family. What changes is everything after the words come out. Whisper on Mac walks through each of the steps.
Possible, with one trade-off worth knowing before you do it.
If you already have an SRT or WebVTT from another tool, drop it on the suubbs player or use Import. The cues keep their start and end times, and you can restyle, edit, retime and export them like any others. But a plain subtitle file only records when each line starts and ends, so the times of individual words inside it are interpolated — suubbs says so when it imports. That is fine for ordinary subtitles. For a word-by-word highlight, generate the subtitles in suubbs instead, so every word carries a measured time.
The one format that brings word timings with it is suubbs' own word-level JSON, which re-imports losslessly, with its styling.
[captions, not transcripts]
…you mainly need transcripts rather than captions, you want a plain-text document out, you work through many files at once, or you work in a language outside our sixteen. Those are real reasons, and its own site will tell you which of its versions covers them.
Choose suubbs if the output is going on a video: word-level timing, readable line breaks, caption styling and an ASS export that matches its preview are the things it is built for, and it is $9.99. Try the free browser version on the first 60 seconds of your own footage before deciding.
Mostly from people who already have a Whisper app.
We have not run a side-by-side test, and both are built on Whisper, so the words should be broadly similar with a comparable model. Where suubbs differs is the timing of each word, measured at about 3 ms, and everything done with the words afterwards.
No. It exports timed subtitle formats — SRT, WebVTT, ASS and word-level JSON. If a document is what you want, a transcription app is the better tool.
No. The Mac app transcribes with whisper.cpp on your own machine, and the free browser version runs its speech model inside the tab. The Mac app needs the network to download its models once and to check for updates; your files never leave the Mac.
Yes — SRT and WebVTT import, keeping each line's start and end. Word timings inside each line are interpolated, so for a word-by-word highlight it is better to generate the subtitles in suubbs.
Large v3 Turbo, quantised, by default in the Mac app, through whisper.cpp. Base and Small English models are offered for speed. The free browser version starts on Base and can load Small or Large v3 Turbo.
No, one file at a time. Your videos stay in the Mac app's library, so you can move between projects, but each is transcribed separately.
Yes. The Mac app runs on macOS 11 or later, on Intel or Apple Silicon. There is no Windows version yet; the browser trial runs anywhere.
Updated November 2026. Everything said here about suubbs comes from its own source. Apps like MacWhisper change their features and pricing, so we have left out any detail we could not verify; check its own site for those. If something here is wrong, tell us at hello@suubbs.com and we will fix it — an inaccurate comparison is not worth a sale.