Offline speech to text on a Mac
You can dictate and transcribe on an Apple-silicon Mac with Wi-Fi off. This page lists what works offline in Vocco, how large the speech models are, and the few things that still go online.
Short answer: Vocco does its transcription on the Mac itself. Hold right ⌘ in any app, speak, and let go: the text is pasted at your cursor without a network connection. Audio and video files can be transcribed offline too, with speaker labels and timestamps.
You need the internet once per speech model, to download it. You don't need an account.
What works with Wi-Fi off
- Push-to-talk dictation in any app. Mail, Slack, Notes, Cursor, Terminal, a browser text box. Vocco puts the text on the clipboard and pastes it at the cursor, so it works anywhere you can paste.
- Real-time transcription. A live transcript window for meetings, lectures or thinking out loud. On-device voice-activity detection splits it into punctuated sentences as you go.
- File transcription. MP3, WAV, M4A, MP4, MOV and the other audio and video formats macOS can read. You get a timestamp on every segment.
- Speaker diarization (Pro, and the first 14 days). Recordings come back split by speaker. The two diarization models (45 MB together) ship inside the app. On long recordings, the current version (1.0.8) can split one person into several speakers; a fix is coming. How it works.
- SRT export and history. Export subtitles, or search the local history. Dictations are kept for 7, 30 or 90 days, your choice; real-time and file transcripts stay until you delete them.
The engines and their download sizes
Apple Speech is built into macOS and works immediately. Qwen3 ASR downloads by itself the first time you open Vocco; the others are one-click downloads in Settings → Models.
| Engine | Download | Good at |
|---|---|---|
| Apple Speech | built in | Starts instantly; one language at a time: your Mac's system language, or the app's language (Chinese or English) |
| Qwen3 ASR 0.6B (int8, Vocco's default download) | 987 MB | Mixed Chinese–English in one sentence; detects the language by itself, including Japanese, Korean and French |
| SenseVoice Small | 938 MB | Fast; Mandarin, Cantonese and English |
| Paraformer Large | 823 MB | Mandarin-heavy speech; the most accurate Chinese of Vocco's engines in our test, and slightly ahead on mixed sentences. Adds a 294 MB punctuation model, downloaded automatically when you pick it |
| Whisper large-v3-turbo | 1.64 GB | Chinese or English sessions; the engine many Mac dictation apps use |
| Whisper Base | 151 MB | The smallest download, for when disk space is tight |
Sizes are what the app downloads, from the model catalog in Vocco 1.0.8. The downloads come straight from public model repositories, not from our servers.
What still goes online
None of these requests carry audio or text.
- Trial and license check at launch and about once a day. No account is involved: the trial is counted per Mac. If you are offline, the app keeps using the last status it knew, so you can still dictate.
- Sign-in, only if you choose to: with Google or Apple, to use a purchase on another Mac.
- Prices for your country, fetched at launch so the in-app plans are up to date.
- Model downloads, once per engine.
- Update checks for new versions.
Apple Speech stays on the Mac too. Vocco asks macOS for on-device recognition only; if macOS can't recognize your language on this Mac, Vocco stops and tells you instead of letting macOS send the audio to Apple's servers. The privacy policy lists exactly what the app and the website send.
How to check it yourself
- Open Vocco once while you are online: it downloads Qwen3 ASR (987 MB) in the background and makes it the default engine.
- Turn off Wi-Fi, or switch on airplane mode.
- Hold right ⌘ in any text field, say a sentence, release. The text appears.
For a stricter test, run a network monitor such as Little Snitch or LuLu while you dictate and watch Vocco's connections stay at zero.
Limits worth knowing
- Apple silicon and macOS 14 or later only.
- The first download is large, about 1 GB per engine.
- Chinese output is in Simplified characters. There is no Traditional Chinese output yet.
- Whisper in Vocco runs one language per session, Chinese or English. For speech that switches language, use Qwen3 ASR.
Questions
Does Vocco send my audio anywhere?
No. Audio is transcribed on your Mac and is never uploaded to us. That includes Apple's own engine: Vocco only lets Apple Speech recognize on-device, and if macOS cannot do that for your language, Vocco stops with an error instead of letting macOS send the audio to Apple.
How much disk space do I need?
About 1 GB for one downloadable engine, such as Qwen3 ASR at 987 MB. Apple Speech needs no download. Whisper large-v3-turbo is the largest at 1.64 GB.
Does it work on Intel Macs?
No. Vocco needs an Apple-silicon Mac (M1 or later) running macOS 14 or later.
Can I use it on a plane?
Yes, once you have downloaded an engine. Dictation, real-time transcription and file transcription all work in airplane mode.
How this page was made: product details come from Vocco 1.0.5's code and model catalog; accuracy claims link to our published benchmark. Drafted with AI assistance and checked against those sources.
Speech to text that works offline
Download a model once and it never needs the internet again. Free, no sign-up.
macOS 14 or later · Apple silicon