Transcription
Available onWeb · beta, capable desktopsmacOSWindowsiOSAndroid
Scribe0’s core job is turning speech into text. On every native app this runs on your device using the Parakeet speech model: audio is processed locally and never uploaded. The web app can do the same on capable desktops, running Parakeet right in your browser (beta), or stream to a cloud service for the highest accuracy: your choice in Settings.
Live transcription
Section titled “Live transcription”Text appears while you speak, so you can keep an eye on the transcript during the consult. Behind the scenes the app shows a rolling partial transcript that is replaced by a final one whenever you pause.
Audio file import
Section titled “Audio file import”Available onWebWindowsiOS · voice memosAndroid
Import an existing recording (WAV, MP3, M4A, and more) and Scribe0 transcribes it like a live session, useful for voice memos or recordings made on another device. Long recordings are handled automatically; a 30-minute consult transcribes in well under a minute on a typical modern PC or Mac.
Speaker identification (diarization)
Section titled “Speaker identification (diarization)”Available onWeb · turn detectionmacOSWindows · on importiOSAndroid · on import
Turn on diarization to label who said what (“Speaker 1 / Speaker 2”), useful for multi-party consults and case conferences.
Platform specifics
Section titled “Platform specifics”- Model: Parakeet 0.6B v2 (English), running locally via sherpa-onnx.
- A smaller Fast model (~126 MB) is available in Settings → Transcription for older or low-power PCs.
- Live transcription is always on and shows text as you speak; file import is Ctrl+I; diarization is a toolbar toggle applied to imports.
- A Diagnostics panel on the Workbench shows live transcription speed, CPU, and memory.
- Model: Parakeet 0.6B v2 (English) on the Apple Neural Engine, fast and battery-friendly on Apple silicon.
- Live transcription and speaker diarization are built in.
- Requires macOS 14+ on Apple silicon (M1 or later).
- Model: Parakeet 110M (English) on the A-series Neural Engine, a smaller model tuned for iPhone.
- Supports voice-memo import and diarization.
- Requires iPhone 13 or later on iOS 26.4+. Free on the App Store.
- Model: Parakeet 110M (English) via sherpa-onnx, running on the CPU (Android’s NPU support is too fragmented across vendors to rely on).
- Live transcription, plus import of an existing recording, either from the file picker or by sharing audio into Scribe0 from another app.
- Speaker labels apply to imported audio, not to live recording.
- Requires Android 7.0+ on a 64-bit phone; currently an invite-only Play alpha.
- Two engines, chosen in Settings: Cloud (AssemblyAI) streams for the highest accuracy, or on-device (Parakeet, beta) runs in the browser via WebAssembly so audio never leaves the page.
- The on-device engine is CPU-based and aimed at capable desktops (about 8 GB of RAM and a recent multi-core processor); on lighter machines and on phones, use Cloud. Note drafting still uses a cloud model.
- Realtime partial/final transcripts as you dictate, either way.