Voice-controlled Ordering System Development -- 2

Freelance — Real-Time Streaming STT (Web)

Existing prototype, deployed and tested in real conditions. Not a rebuild — we need it hardened.

What runs today: a commercial streaming STT API over WebSocket (EU-hosted, provider chosen — named in private exchange), wrapped in a shared JS module. Mic capture and PCM 16 kHz streaming via Web Audio API, custom vocabulary, free-form French dictation parsed into structured lines and matched against an internal catalogue. Vanilla JavaScript, no framework, no build step; Firebase Realtime Database; PWA on smartphone.

What we need:

Reliability in noisy conditions — false finals on silence reach the application layer. Endpointing, thresholds, confidence handling.
Securing the API key — currently client-side, needs a serverless function issuing session URLs.
Usage and cost control — real metering and per-user limits.
ScriptProcessor → AudioWorklet, validated on iOS Safari and Android Chrome.
Custom vocabulary tuning from the failure transcripts we already log.

Profile: demonstrable experience integrating a real-time streaming STT API over WebSocket — tell us which providers you have shipped with. Solid vanilla JS and Web Audio API. Firebase RTDB. French-language voice interfaces. Willing to work incrementally on an existing codebase.

Please do not propose local/self-hosted engines or switching provider. Both decisions were made after testing.

Practical: short engagement, codebase shared on selection. Propose timeline and budget broken down per item — partial proposals welcome.