Configuration

Server CLI

cargo run --release -- --help
# or
docker run lutgaru/sinsajo-server:latest --help

Arg	Default	Description
--model	prompt	ParakeetTDT / Canary180M
--autodownload-model	false	Download if missing
--port	8765	WebSocket port
--model-dir	./models	Model storage
--records-dir	./records	Recordings output

Env / file: sinsajo-config.json persists last chosen model.

Client Settings (in-app)

• Server IP — kWsUrl default; overridden in Settings UI
• Target language — en/es/fr/de/pt (enabled per model_info.languages)
• Save audio — toggle + format wav/ogg
• Microphone gain / source

Code default:

const String kWsUrl = 'ws://192.168.1.100:8765';

VAD Tuning

In audio_service.dart via VadHandler.startListening:

positiveSpeechThreshold: 0.5, // start
negativeSpeechThreshold: 0.35, // stop
redemptionFrames: 8,          // silence to end utterance
preSpeechPadFrames: 1,
minSpeechFrames: 3,
frameSamples: 512, // 32ms

Try: 0.3 / 0.2 / 12 for quiet speakers or long pauses.

Audio recording

Client sends with start:

{"type":"start","sample_rate":16000,"save_audio":true,"format":"wav","target_language":"es"}

Server saves on clean to records/YYYY-MM-DD_HH-mm-ss.wav.

GitHub Pages deployment (docs-site)

Config in astro.config.mjs:

site: 'https://lutgaru.github.io',
base: '/Sinsajo/',

If you add a custom domain, set base: '/' and add public/CNAME.

Flutter Web must use:

flutter build web --base-href /Sinsajo/app/