Voice I/O for the chat panel (deferred) #56

New issue

Open

opened 2026-06-14 19:52:25 +00:00 by james · 0 comments

james commented

2026-06-14 19:52:25 +00:00

Owner

Layer voice input and output onto the chat panel. Voice was deferred from the first cut of the agent-interaction epic so the text surface could stabilise first; this ticket picks it up afterward.

Scope (revisit when starting)

STT: browser-native (Web Speech API) first cut. Allowance to swap to server-side Whisper later as a self-host-friendly alternative.
TTS: browser-native voice for the first cut.
UX: push-to-talk button in the chat panel. Transcript shown as the user speaks; submission happens on release.
Settings: voice on/off per device.
Out of scope for the first cut: continuous always-listening, voice activity detection, anything that requires server-side audio processing.

Acceptance criteria

A user can press and hold the mic button, speak, and see their words become the next chat message.
The agent's reply is read aloud through the browser's default voice when voice TTS is enabled in settings.
Voice can be turned off entirely from settings without affecting text chat.

Part of epic #47. Depends on the Built-in chat UI. Deferred — not blocking the epic's exit criteria.

Layer voice input and output onto the chat panel. Voice was deferred from the first cut of the agent-interaction epic so the text surface could stabilise first; this ticket picks it up afterward. ## Scope (revisit when starting) - STT: browser-native (Web Speech API) first cut. Allowance to swap to server-side Whisper later as a self-host-friendly alternative. - TTS: browser-native voice for the first cut. - UX: push-to-talk button in the chat panel. Transcript shown as the user speaks; submission happens on release. - Settings: voice on/off per device. - Out of scope for the first cut: continuous always-listening, voice activity detection, anything that requires server-side audio processing. ## Acceptance criteria - [ ] A user can press and hold the mic button, speak, and see their words become the next chat message. - [ ] The agent's reply is read aloud through the browser's default voice when voice TTS is enabled in settings. - [ ] Voice can be turned off entirely from settings without affecting text chat. Part of epic #47. Depends on the Built-in chat UI. **Deferred — not blocking the epic's exit criteria.**

james added the

area:pwa

label

2026-06-14 19:52:54 +00:00