FE-1749: Add Brunch voice selection, previews, and device routing

[kostandinang]

Note

Depends on #9774, which contains the generic Petrinaut voice-audio controls. This PR contains the Brunch and website implementation.

🌟 What is the purpose of this PR?

Implement extended Brunch/Petrinaut Voice settings without crowding the conversation. Keep microphone mute separate from app-output mute and volume, and provide provider-specific voice selection, static previews, browser-device routing, and persistence through Petrinaut's Audio options.

🔍 What does this change?

• Persist provider-specific voice choices and validate them at both session-creation endpoints.
• Preview selected voices from versioned static assets without provider requests or microphone audio.
• Route microphones with getUserMedia/replaceTrack and speakers with setSinkId/selectAudioOutput where supported.
• Recover disconnected devices to system defaults while preserving mute and volume state.
• Forward Realtime speaking speed for the next response without interrupting current playback.
• Keep existing Brunch conversation and transcription behavior unchanged.

🔗 Related links

• FE-1749: Add extended Brunch voice audio settings (internal).
• Petrinaut foundation: FE-1749: Support Brunch voice audio settings in Petrinaut #9774.
• Voice UI discussion (internal).

🚫 Blocked by

• Merge FE-1749: Support Brunch voice audio settings in Petrinaut #9774.
• Complete physical-device and spoken-conversation validation for Live and Realtime before marking ready.

⚠️ Known issues

• Physical-device routing, permission rejection, unplug/reconnect, and spoken-conversation behavior remain unverified on real hardware.
• Output-device selection depends on browser support and permission. Numeric speaking speed is Realtime-only.
• Local tests use NODE_OPTIONS=--no-experimental-webstorage because Node 26's built-in storage otherwise overrides jsdom storage.

🛡 What tests cover this?

• 227 focused website tests covering provider forwarding, voice persistence and previews, device routing and recovery, permissions, concurrency, and both Voice providers.
• @apps/petrinaut-website TypeScript checks.
• The original combined PR passed the full CI suite before this reviewer-oriented split.

❓ How to test this?

NODE_OPTIONS=--no-experimental-webstorage yarn workspace @apps/petrinaut-website test:unit \
  src/main/app/voice-interview/live-conversation-control.test.tsx \
  src/main/app/voice-interview/live-conversation.test.ts \
  src/main/app/voice-interview/openai-realtime-session.test.ts \
  src/main/app/voice-interview/voice-audio-settings.test.ts \
  src/main/app/voice-interview/voice-audio-settings/preview.test.ts \
  src/main/app/voice-interview/voice-interview-control.test.tsx \
  src/server/voice/openai-live-session.test.ts \
  src/server/voice/openai-realtime-call.test.ts
yarn workspace @apps/petrinaut-website lint:tsc

For manual validation, configure Brunch Voice, start a Live session, and open Audio options. Mute the microphone before selecting a voice preview; verify output mute/volume, device selection and fallback, status visibility, and next-session voice application. Repeat with the Realtime provider and verify that speed changes affect the next response.

Pre-Merge Checklist 🚀

• This consumer PR does not directly modify a publishable library.
• User-facing documentation and the Petrinaut changeset are in FE-1749: Support Brunch voice audio settings in Petrinaut #9774.
• No execution-graph changes are required.

📹 Demo

live-demo.mp4

Live mode mixer controls demo

live-mixer-controls-demo.mp4

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
hash	Ready	Preview	Sep 18, 2026 6:29pm UTC
petrinaut	Ready	Preview	Sep 18, 2026 6:29pm UTC
petrinaut-docs	Ready	Preview	Sep 18, 2026 6:29pm UTC

1 Skipped Deployment

Project	Deployment	Actions	Updated
hashdotdesign-tokens	Ignored	Preview	Sep 18, 2026 6:29pm UTC

[CLAassistant]

All committers have signed the CLA.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 8cabbd0. Configure here.}

[cursor]

Meter dies after microphone switch

Medium Severity

replaceMicrophone calls #syncMicrophoneTrack before destroying the meter, then rebuilds the AudioContext without starting sampling again. After a device switch while the mic is live, #startMeter never runs, so microphoneLevel stays at 0 and the dock waveform freezes until a later turn-state change.

 

^{Reviewed by Cursor Bugbot for commit 8cabbd0. Configure here.}