Echos

On-device voice notes + transcription built with Expo / React Native.

What it does

• Record + transcribe: real-time streaming and file-based transcription via sherpa-onnx
• On-device storage: sessions + transcripts stored locally, audio encrypted at rest (AES-GCM)
• Settings: theme, language, ASR model, per-model language, incognito mode, and advanced options

Quickstart

npm install
npm start

This app uses native modules and requires a development build — Expo Go will not work:

npm run ios
# or
npm run android

Design System Gallery

Longpress the Echos wordmark on the homescreen in dev mode to open the design system gallery.

Testing

npm run test              # run all tests
npm run test:coverage     # run with coverage (thresholds: 95/90/95/95)

Models

Transcription runs on-device via sherpa-onnx. Only Whisper Tiny is bundled; the rest are downloaded on demand from HuggingFace into DocumentDirectory/models/.

Model	Size	Languages	Modes	Bundled
Whisper Tiny	~104MB	99	realtime + file	yes
Whisper Base	~161MB	99	realtime + file	no
Whisper Small	~375MB	99	realtime + file	no
NeMo Parakeet V3	~670MB	25 European	realtime + file	no
Qwen3 ASR	~983MB	30 + 22 Chinese dialects	file only	no