Speak. Watch it type.
Echo is a real-time speech-to-text app that runs entirely on your Mac.
Streaming drafts appear in milliseconds, a large model finalizes on pause — all offline, and your audio never leaves your computer.
Why Echo
Privacy-first on-device recognition, with a transcription experience that doesn’t compromise.
Fully on-device recognition
The recognition engine runs entirely offline — no cloud uploads, no account. Your voice stays yours.
Streaming word by word
Text appears as you speak with millisecond-latency drafts; the main model finalizes after each pause for accuracy.
Two audio sources
Microphone (with selectable input device), or system audio — live captions for meetings, classes, and videos.
Smart segmentation & speakers
Sentences are split by pauses and paragraphs; a voice-print model labels speakers A B C D E.
History & export
Every recording is saved automatically. Review and copy anytime, or export to .txt transcripts and .srt subtitles.
Optional AI rewrite
Off by default. Configure any OpenAI-compatible endpoint (keys stay in the system keychain) and turn a full transcript into a polished document in one click.
Two engines: fast and accurate
A streaming engine drafts in milliseconds while a large model finalizes — both run on your Mac.
Audio stays on your machine
Microphone or system audio is captured at 16 kHz; energy-based VAD detects utterance boundaries — all data stays on the device.
Millisecond word-by-word
The streaming Zipformer engine writes as you talk — light-gray draft text appears on screen within milliseconds.
Finalized by Qwen3-ASR
After a pause of about 0.8 s, Qwen3-ASR 0.6B finalizes: automatic punctuation and number normalization turn the draft into clean text.
Want maximum responsiveness? Switch to Turbo mode to skip finalization — the draft is the final text.
Recognition models download automatically on first launch (fully offline afterwards), with integrity checks and no user identifiers.
Where Echo shines
From meeting notes to dictation — one Mac does it all.
Meetings & interviews
Turn on system-audio capture for a complete record of every call; speaker labels keep everyone’s contributions clear.
System audio · Speakers A–EWriting & dictation
Speak into the microphone and pause to form sentences; optional AI rewriting turns speech into polished prose.
Microphone · Optional AI rewriteVideos & online classes
Add live captions to locally played videos; export .srt subtitles or a transcript afterwards to review.
System audio · Export .srtPrivacy isn’t a feature. It’s the foundation.
Echo’s App Store privacy label reads “Data Not Collected”: no cloud uploads, no account system, and no analytics or advertising components.
- All voice audio is recognized on your device — never uploaded
- Transcripts are stored locally only, fully under your control
- Only two network calls exist: the first-run model download, and AI rewriting you explicitly enable (sent directly to the service you choose)
App Store privacy label
Data Not Collected
The developer does not collect any data from this app.
Get started in four steps
No sign-up. Open it and speak.
-
1
Launch
Echo downloads its recognition models on first launch. Everything works offline afterwards.
-
2
Pick a source
Choose the microphone or system audio (the latter needs screen-recording permission in System Settings).
-
3
Speak
Press start and watch words appear as you talk; text finalizes automatically when you pause.
-
4
Review & export
Recordings land in history automatically — copy or export to .txt / .srt anytime.
Turn your voice into text — safely
No account · Data never leaves your device · macOS 13 or later