Android

Platform details for Scribe0 on Android (Kotlin + Jetpack Compose, Android 7.0+, 64-bit phones, currently an invite-only Google Play alpha). For installation and first steps, see Get started on Android.

The speech model


Model	Parakeet 110M (English)
Runtime	sherpa-onnx (ONNX), on the CPU
Download	~130 MB, once, on first launch

The model is fetched from downloads.scribe0.com on first launch and stored in the app’s private storage, so transcription works offline from then on. It runs on the CPU: Android’s NPU story is fragmented across vendors, so Scribe0 targets the one processor that behaves the same on every phone. Transcription is comfortably faster than real time on a modern handset.

Live transcription shows a rolling partial transcript that is replaced by a final one each time you pause.

Audio import and speaker labels

Import an existing recording (m4a, mp3, wav, ogg, and more) from the in-app picker, or share audio into Scribe0 from another app. Imports can be transcribed with speaker labels (“Speaker 1 / Speaker 2”), which downloads a second set of models (~46 MB) the first time you use it.

Note drafting

Enhance sends the transcript text (never the audio) to a sovereign Amazon Bedrock endpoint in Australia, where inference runs and stays. The app asks for your consent once, before the first cloud call.

It works with no sign-up on a free daily allowance, and the app shows how many enhances you have left. Pasting a Scribe0 Cloud key in Settings lifts the cap and unlocks the larger models. Each enhance you run adds a tab, so you can re-run a transcript with a different prompt and compare the drafts side by side.

Prompts, history, and diagnostics

Prompt library: four built-in styles, plus create, edit, and delete your own. Each prompt can pin its own model. You can also browse the shared prompt gallery from inside the app.
History: sessions save automatically, with full-text search, restore, and delete. Stored in a local database that is encrypted at rest with a key held in the Android Keystore.
App lock: optionally require fingerprint, face, or PIN to open the app (Settings → Privacy).
Diagnostics: each session records transcription speed, first-text latency, peak memory, and thermal state. They contain no patient information.

The app requests exactly two permissions: microphone and internet.

Current limitations

No background recording. Scribe0 doesn’t hold a foreground service, so recording pauses when you leave the app, and when a call or a headphone unplug takes audio focus. Keep it on screen during a consult.
Notes aren’t versioned. Only the latest enhanced note is kept per session; re-enhancing a saved session overwrites it. (The original transcript is always kept.)
The transcript isn’t editable in-app. Share it out to edit it.
Share, not copy. Use the share sheet to send a note to another app; there is no clipboard copy button yet.
English only, like the other on-device apps.

Android FAQ

How do I join the alpha?

It’s a closed alpha on Google Play. Request access via scribe0.com, and include the Google account email you use on your phone. Once you’re added as a tester, the Play listing opens for that account and you install it like any other app.

Will it run on my phone?

If it’s a mainstream handset from the last several years running Android 7.0 or later, yes. Scribe0 ships 64-bit (arm64) only, so it won’t install on very old 32-bit phones or on most emulators.

Does it drain the battery?

Transcription runs on the CPU rather than a dedicated accelerator, so it works the phone harder than the iPhone build does. A normal consult is fine; expect some warmth on very long sessions.

Why is the phone model smaller than the desktop one?

The 110M model is sized for phone memory and battery. It’s excellent for clear dictation; the desktops’ 0.6B model has the edge in noisy rooms or with heavy accents.

Where are my notes?

On the phone, in History, in an encrypted local database. No account, no cloud sync. Uninstalling the app deletes them.