Does recognition upload my audio or video?
No. Recognition and editing run on your computer, and SpeechSub has no cloud recognition. The app reads files you choose and saves captions, drafts, and necessary temporary files on that computer.
A macOS app for local speech recognition and caption proofreading
Recognize locally.
Proofread in place.
SpeechSub turns speech into captions on your Mac, then lets you fix them against the waveform.
Recognition runs on your computer. The audio and video you choose are not uploaded to SpeechSub’s servers.
Features
Recognition, proofreading, import, and export all happen on your computer.
Recognition stays on your computer
Choose audio or video and get captions from local Whisper. High-resolution video is recognized from its audio track, and the whole video is not read into memory.
Proofread on the waveform
Edit text on the timeline and find and replace. Your glossary can correct common mistakes after recognition without moving timestamps. Playback from 0.5× to 3× changes listening speed only.
Import, then export
Bring in existing SRT, WebVTT, ASS, or SSA and edit without recognizing again. Export SRT, WebVTT, ASS, or TXT. Named drafts stay on this computer.
Download only when you need to
Proofreading does not require a recognition model. The speech-or-silence file ships with the app: ggml-silero-v6.2.0.bin (885,098 bytes). Optional timing correction is about 1.8 GB, and the size is shown before the download.
Frequently asked questions
The questions that come up in use, answered here.
Do I have to download a model?
Not always. Importing, proofreading, and exporting do not need a recognition model. To recognize new audio or video, download Whisper when you need it: about 78 MB, 252 MB, or 834 MB, with the size shown first. The file that separates speech from silence ships with the app, ggml-silero-v6.2.0.bin (885,098 bytes), so you do not download it separately.
What is “Recognize and correct timing”?
An optional step. Whisper writes the captions, then Qwen aligns the timing. It does not turn speech into text. That model is about 1.8 GB, downloads only when you need it, and shows its size first. If it is not on your computer yet, the app asks you to download it instead of finishing on its own.
Why doesn’t the Apple Speech switch in Settings work?
It does not work yet. Apple Speech is a system capability. It does not appear in model management, and you do not download it as a recognition model. Settings may show the switch, sometimes only on macOS 26 or later, but it is still a preview and is not connected to recognition. Recognition stays on local Whisper.
Can I proofread captions I already have?
Yes. Import SRT, WebVTT, ASS, or SSA, then preview, edit, and export on your computer.
Which formats can I export?
Edited captions export as SRT, WebVTT, ASS, or TXT.
How do I get support?
Email 6666699999@live.com. Do not attach passwords, keys, complete media files, or unredacted transcripts. This site does not offer a download or a purchase link.
SpeechSub and TalkTrim are macOS tools from the same editing family. TalkTrim suggests reviewable rough cuts for Final Cut Pro spoken-word projects. SpeechSub handles local recognition and caption proofreading. TalkTrim website