PROTOTYPE v13 · dark research-grade UI · real LLM at Reframe · deterministic fallback · debug via ?debug=1
Inner Voice Coach

Inner Voice Coach

A brief, guided space for the moments your own thoughts talk you out of asking for help. You can speak to it or type, whichever feels easier.

your participant ID

This is the same ID you entered on the consent form. We use it only to match your responses across the surveys and this artifact.

talk to your coach speak or type
On by default — you speak, the coach listens and speaks back. You can always type instead, and a mute button at the top silences the coach's voice (useful in shared or quiet spaces). Turn this off to use text only.

About this artifact
  • This is a research artifact, not therapy or crisis care.
  • Please do not enter names, addresses, phone numbers, or other identifying details.
  • In this prototype, your responses stay in this browser session. In the deployed study, limited text needed for the rewrite may be processed through the PSU-approved research system. Please do not enter names or identifying details.
  • You can pause at any time using the settle button at the top.
notice
reframe
act
carry
your space

Inner Voice Coach

A brief, guided space for the moments your own thoughts talk you out of reaching for help. You choose where to start, and whether to speak or type. Tap below to talk it through, and the tools keep what you find so it is here when you need it again.

quick tools

Mantras

Short phrases for when the inner voice gets loud.

Your mantras will appear here.

Small Steps

Tick a tiny help-seeking action you could do today.

Reach Out

PSU resources and crisis lines for when you need them.

Think of one trusted person. Hold their name in your head — we will not ask you to enter it here. When you leave this artifact, consider sending them a short message.

Not therapy. Crisis: call/text 988 · text HOME to 741741 · CAPS 814-863-0395
state machine
v14 voice-coach demo · what is real
implemented: hub-and-spoke home, real voice input (Web Speech API) with live transcription, coach text-to-speech speak-back + mute, breathing coach orb that reacts to listen/speak, modality telemetry (voice/text/voice-then-edit), flow logic, safety-pipeline state, Burns picker, 3-point believability anchors, Settle, crisis-flag pause.
LLM: calls /api/reframe Cloudflare Worker if reachable; falls back to deterministic candidates if the Worker fails, returns invalid JSON, or safety-rejects. The logged mode field shows which path ran.
voice notes: deployed, voice records audio and transcribes server-side via Cloudflare Whisper (uniform across Chrome, Safari, iOS, Firefox; audio not retained). Opened directly from disk for local testing, it falls back to the browser speech API (Chrome). Spoken transcripts run the same crisis check as typed text before any reframe.
planned, not yet live: Firestore event logging (currently client-side only, in-session), Qualtrics post-survey URL (placeholder), Whisper server-side transcription (the deployed study can route audio through Cloudflare Workers AI Whisper instead of the browser API).