Expertise Needed for Web Speech API Optimization

JS Freelancer needed — Browser-based Voice Recognition (Web Speech API)

Context:
I'm developing a PWA web application (vanilla HTML/CSS/JS, Firebase). A "hands-free" mode allows users to validate items by voice, without touching the screen.

The problem:
Web Speech API on Android Chrome is unstable in noisy environments:
- Sessions die silently after a period of silence
- Short trigger words ("hop", "ok", "next") frequently missed or misrecognized
- Erratic behavior across Android versions
- Conflicts between speech synthesis (TTS) and recognition (STT)

What I already have:
- Working voice mode with short sessions + automatic restart
- interimResults, maxAlternatives, expanded trigger word list
- Accent stripping in transcript comparison

What I'm looking for:
Someone who has already solved these issues in production — not theory. Ideally with one of the following approaches:
- Advanced Web Speech API optimization (VAD, fine-grained state management, heuristics)
- Whisper integration via Transformers.js in a mobile context
- Any other proven browser-side approach, no server required

Stack: vanilla HTML/JS, Android Chrome, Firebase Realtime Database. No framework.
Please apply with references from similar cases.