JapaCounter

JapaCounter is a production-grade, privacy-first, AI-assisted mantra repetition counter built as a precision spiritual instrument.

Vision

• Trustworthy counting over flashy novelty.
• Immersive premium minimalist ritual UX.
• Local-first audio processing and storage.

Stack

• Next.js App Router + React + TypeScript strict mode
• Tailwind CSS + Framer Motion
• Vercel-compatible architecture

Architecture

• app/: routes for marketing, live session, settings, history, privacy.
• hooks/useChantSession.ts: microphone pipeline and real-time loop.
• lib/audio/fingerprint.ts: deterministic fingerprint extraction + cosine similarity.
• lib/engine/counting-engine.ts: anti-double-counting, cooldown gating, confidence thresholds.
• hooks/useLocalSessions.ts: local-only persistence.

Audio Pipeline

1. Capture microphone stream with getUserMedia.
2. Analyze frames with AnalyserNode.
3. Build compact spectral-temporal fingerprint per frame.
4. Compute similarity against recorded reference mantra.
5. Run VAD energy gate to filter silence/breathing/noise.
6. Send candidates to deterministic counting engine.

Counting Philosophy

• Deterministic first: a repetition is counted only when threshold and cooldown windows pass.
• Adaptive cooldown scales with the reference tempo.
• Confidence is continuously exposed to UI.

Privacy

• No server-side audio storage.
• No mandatory account.
• Session history kept in localStorage.

Deployment

npm install
npm run build

Deploy directly to Vercel.

Roadmap

• AudioWorklet + Worker offloading
• Silero/WebRTC VAD integration
• Optional ONNX local model refinement
• IndexedDB archival + export/import
• PWA offline temple mode