Utterlane：Android 离线语音转录应用

Utterlane - Blazingly fast accurate offline transcription for Android.

Get started · Features · Privacy · Documentation · Contribute

Utterlane is a free, open-source **offline voice-typing and audio-transcription app for Android**. Dictate into any app, use your keyboard's microphone button, or share a voice memo to the app to transcribe. No account or subscription is required.

Features

Offline and private: speech recognition runs on your device.
Streaming text as you speak: receive completed speech segments while recording continues.
Blazingly fast: When you stop speaking, the transcription is already done. That's because of clever optimizations, one of which being to start transcribing in the background, and careful memory handling between transcription and audio recording allows to concurrently do both with no performance hit.
SOTA accurate models: Moondream Parakeet Ultra (Sept 2026, ~670MB) is the default model. For slow devices, Moondream Parakeet Redux with ternary quantization (~160 MB) is available. For more advanced phones, lots of custom SOTA models can be imported via CrispASR (eg, R2T2).
Reliable, robust: Hardened recording and transcription that leaves no room for failure. Worst case scenario, voice recordings history with autopruning allows you to always recover from a failed transcription.
Works across apps: use a keyboard mic, accessibility button, or floating mic.
25 languages: language detection is automatic, even when you switch languages live in the middle of a recording.
Audio-file transcription: share or open vocal memos and recordings, or monitor a folder for new audio.
Personalize your results: add word corrections and optional speaker labels.
Real-time meetings transcripts: Optional streaming speakers diarization with auto detection of the speakers count allows to transcribe meetings in real-time.

Get started

Install

You need Android 8.0+ with ARM64 support and space for a speech model (approximately 159–674 MB).

Download the Utterlane-<version>-arm64-v8a.apk from GitHub Releases and open it to install. If prompted, allow installation from your browser or file manager. The release APK will be available after the first signed release is published; until then, see building from source.

You can also track releases with Obtainium using the button above, or find lrq3000/Utterlane in Komi Store.

Store availability

F-Droid and a free Google Play listing are planned. See the installation guide for update options and advice on switching between distribution channels.

Your first dictation

Open Utterlane and follow the setup guide. Choose a recommended speech model and download it, or import existing model files. After setup, recognition works offline. The guide includes optional in-app dictation and audio-sharing tests and can be reopened from Settings.
Grant Microphone access.
Choose an input method:
- Keyboard mic: open Keyboard Integration → Voice Input Method, enable Utterlane, and enable the voice-input key in a compatible keyboard such as HeliBoard.
- Accessibility button: enable Utterlane's text-input accessibility service in Android Settings.
- Floating mic: allow Display over other apps and enable the floating button in Utterlane.
Focus a text field, tap the microphone, and speak. Tap the recording panel to finish. If text cannot be inserted, use the clipboard or transcript export.

Have an audio file? Share it with Utterlane or choose Open with → Utterlane in your file manager.

Privacy

Audio and transcripts are not uploaded to a speech service. No analytics or account is required.
Network access is used for requested model downloads; local model import is also available.
Recording history defaults to one-hour retention. You can disable it or change retention in Settings; existing choices are preserved. Saved audio stays private to the app and is excluded from Android cloud backup and device transfer.
Exporting or copying a transcript shares it with the receiving app.

Read the privacy policy for permissions and retention details.

Documentation

Core values: the product principles guiding responsiveness, speed, reliability, privacy, and user ownership.
User guide: installation options, models, languages, speaker labels, recording history, and settings.
Developer guide: building, testing, architecture, streaming, and model internals.
Release and store submission guide
QA reports · Brand artwork

Contributing

Bug reports, translations, accessibility feedback, and code are welcome. Open an issue with your app version, device, selected model, and steps to reproduce the problem. Remove private speech and transcript content from shared logs.

See CONTRIBUTING.md. AI-assisted contributions are welcome when humans sanity check the outputs before submission.

License and acknowledgments

Licensed under Apache-2.0. Upstream copyright notices are retained; Utterlane contributions are copyright 2026 Stephen Karl Larroque and Utterlane contributors. Models and third-party libraries retain their own licenses; see dependency acknowledgments.

Lineage and maintenance

Utterlane is an independently maintained fork of TranSlander, originally developed by hatsch and its contributors with credited assistance from Claude Code.

Maintained by Stephen Karl Larroque (LRQ3000@GMAIL.COM), with a focus on responsive, local-first transcription. It installs separately from its predecessor; settings and recordings do not migrate automatically.