Voice transcription that never phones home

What it does NotelyVoice is a cross-platform voice transcription and note-taking app built with Compose Multiplatform for Android and iOS. It records audio, transcribes speech to text using OpenAI Whisper locally on the device, and organizes the results into a rich-text notebook with folders, tags, and search. The open-source version in this repo is GPL-3.0 and distributed free on F-Droid, while a separate proprietary rebuild with a subscription model lives on Google Play and funds development of both.

The interesting bit The app handles long recordings without choking by streaming audio through a chunked decoder: it breaks files into 30-second overlapping segments instead of loading everything into memory at once. There is also a custom rich-text editor built from scratch without third-party libraries, and the whole UI layer shares a single Compose Multiplatform codebase across mobile operating systems.

Key highlights

• Runs Whisper entirely offline after the model downloads; no cloud transcription.
• Memory-efficient processing uses overlapping 30-second chunks with a 3-second overlap to avoid OOM errors on large files.
• Supports multilingual speech recognition (the README cites 50+ languages).
• Clean architecture split into UI, Presentation, Domain, and Data layers using Koin, ViewModels, and Kotlin Coroutines.
• Native Compose Navigation and a from-scratch text editor keep external dependencies minimal.

Caveats

• The open-source F-Droid version carries the original UI; the redesigned interface is reserved for the proprietary Google Play subscription app.
• Model sizes range from 39 MB to 1.5 GB, so storage and performance will vary significantly by device and chosen model.

Verdict Worth a look if you need private, offline transcription on a phone and prefer open-source tooling. Skip it if you want the polished, redesigned experience, since that lives in the separate paid app.