Offline speech-to-text for Mac.
Drop in a meeting recording and get a transcript split by speaker, with a time on every sentence, in about 8 seconds. Your audio never leaves your Mac. No account, no cost.
Requires a Mac with Apple silicon and macOS 26 or later. Speed measured in Fast mode on an M4 Pro.
Unplug the internet. It still works.
Recognition, speaker detection, translation, and polishing all run on your Mac. Your audio is never uploaded and there is no account to create. Tested with network access blocked, transcription and speaker detection finished as usual. You only need the internet to download models and language packs the first time.
An 8-minute recording in 8 seconds.
Measured on an M4 Pro with an 8-minute, two-person conversation: 7.5 seconds in Fast mode and 73 seconds in Accurate mode, with both speakers identified. Live captions update their draft text about every 0.1 seconds.
One window, four rows of settings.
Drop a recording into the window and click Start. You never sign in, and there is no subscription or cloud setup to deal with.
Built on the engines inside your Mac.
Fast mode uses Apple's speech recognition, translation uses Apple's Translation, polishing uses Apple Intelligence, and speaker detection runs on the Neural Engine. The whole app is a 12 MB download.
I turn a lot of recordings into text, then use it for notes and summaries. With that much audio, I wanted a tool that is fast and keeps the recordings on my own computer, so I built Lisvo. Recognition and speaker detection run on the Mac's built-in speech recognition and Neural Engine, and an 8-minute meeting is done in seconds to just over a minute. It works well for me, and I figured others could use it too, so I made it open source and free for everyone.
Comparison
MacWhisper is a mature Mac transcription app. It can record meetings automatically, handle dictation, and connect to Notion, and if you need those features it is worth the price. If you just want your recordings and online meetings turned into text, here is how they differ:
| Item | Lisvo | MacWhisper |
|---|---|---|
| Price | Free | Free tier; Pro is a one-time €64 |
| Source code | Open (Apache 2.0) | Closed |
| Speech recognition | Apple speech recognition (Fast), Whisper large-v3-turbo (Accurate) | Whisper, Parakeet |
| Speaker detection | pyannote segmentation + WeSpeaker voiceprints, on the Neural Engine | Yes |
| Translation | Apple Translation, on device | Yes |
| AI polish / processing | Apple Intelligence, on device | Your own API keys, or local Ollama and LM Studio |
| Export formats | TXT, timeline, SRT, VTT, JSON, MAXQDA | TXT, Markdown, PDF, HTML, DOCX, SRT, VTT |
MacWhisper details are from its official website (October 2026).
Transcripts
Fast mode uses the Mac's built-in speech recognition, needs no model download, and finishes an 8-minute recording in 8 seconds. Accurate mode uses Whisper large-v3-turbo and detects the language for you. Lisvo can work out how many people are speaking, or you can set the number yourself.
Click any sentence to play just that part of the original audio. Checking a misheard word no longer means dragging the playhead around.
Live captions and translation
Choose Computer audio and anything your Mac is playing becomes captions: Zoom, Teams, YouTube, any app. Choose Microphone for live talks and face-to-face conversations. Captions float above your meeting window, and the draft text updates about every 0.1 seconds.
Captions can be translated into Traditional Chinese, English, Japanese, or Korean. For speakers who mix Chinese and English, choose Chinese–English: Chinese parts are translated into English and English parts into Chinese. Translation also runs locally.
Exports
Each transcription gets its own folder: plain text split by speaker, a timeline with one sentence per line, SRT and VTT subtitles, a JSON file with all the data, and a file you can import straight into MAXQDA, so qualitative research needs no format conversion.
Turn on AI polish for an extra version with typos fixed, punctuation added, and filler words removed. The original transcript and timings stay untouched. Polishing uses Apple Intelligence on your Mac and is not uploaded either.
Weekly sync 2026-10-30 1405
iPhone inbox
Set up a Shortcut once on your iPhone. Share a Voice Memos recording and it is saved to the Lisvo/收件 folder in iCloud Drive. While Lisvo is open on your Mac, it transcribes the file and saves the transcript back to iCloud, ready to read in the Files app. This is the only feature that goes through iCloud. Everything else stays on your Mac.
Lisvo is not notarized by Apple. Notarization requires the paid Apple Developer Program, at US$99 a year, which this project does not pay for, so macOS shows a warning the first time you open it. Many free, open-source Mac apps from independent developers ship this way. Do this once and it opens normally afterward:
These steps follow Apple's guide, Open a Mac app from an unknown developer.
Yes. Lisvo is a free, open-source speech-to-text app for Mac. Recognition, speaker detection, and translation run on your Mac, and once the models are downloaded it works without the internet. It needs a Mac with Apple silicon and macOS 26 or later.
Open Lisvo, drop the recording or video into the window, and click Start. The transcript is split by speaker with a time on every sentence, and saved as plain text, a timeline, SRT, VTT, and JSON. An 8-minute recording takes about 8 seconds in Fast mode on an M4 Pro.
No. Lisvo is free, and its source code is on GitHub under the Apache 2.0 license.
MacWhisper Pro is a one-time €64 purchase and does more, such as automatic meeting recording, dictation, and Notion integration. Lisvo is free and open source, focuses on turning recordings into transcripts and on live captions with translation, and also exports a file you can import into MAXQDA. If transcription is all you need, Lisvo covers it.
Yes. Once the models and language packs are downloaded on first use, recognition, speaker detection, translation, and polishing need no internet. Tested with network access blocked, transcription and speaker detection finished as usual.
No. Recognition, speaker detection, AI polish, and translation all run on your Mac. Only the iPhone inbox feature sends recordings and transcripts through your own iCloud Drive.
Recognition supports Chinese (Taiwan), Cantonese, English, Japanese, and Korean, and Accurate mode can detect the language. Live translation supports Traditional Chinese, English, Japanese, and Korean, plus Chinese–English in both directions.
Fast mode uses the Mac's built-in speech recognition, needs no model download, and shows text the soonest, but you choose the language first. Accurate mode uses Whisper large-v3-turbo, downloads its model on first use, and can detect the language.
Lisvo is not notarized by Apple, so macOS blocks it the first time. Go to System Settings > Privacy & Security, click Open under Security, then Open Anyway. After that it opens normally.
No. Lisvo relies on speech recognition added in macOS 26, and the app is built only for Apple silicon.
A 12 MB download. Requires a Mac with Apple silicon and macOS 26 or later.
