P1 — harden Samsung assistant background voice, audio and reconnect lifecycle #203

Open
opened 2026-08-24 06:24:56 +00:00 by lost-rob0t · 0 comments
lost-rob0t commented 2026-08-24 06:24:56 +00:00 (Migrated from github.com)

Parent: #194
Depends on: #132, #175, #196, #197
Integrates with: #133, #134, #152, #159

Goal

Harden the shared Android voice edge for system-assistant activation, application activation and background invocation on supported Samsung lifecycle paths.

Required behavior

Android owns capture/playback/audio focus and immediate local stop; zara-server owns canonical VAD/STT/routing/LLM/tools/TTS. Support partial/final transcripts, streamed response/audio, barge-in, server cancellation, Bluetooth/headset route changes, UI close/reopen, background/foreground transitions, process recreation, authentication/reconnect and role loss.

Wake-word behavior must follow #195 research and must not assume unrestricted permanent microphone access.

Tests

TDD audio-focus gain/loss, route changes, local stop budget, stale audio rejection, cancellation/replay, permission revocation, UI/process death, daemon/network loss, bounded reconnect, multiple devices/principals and redacted diagnostics. Hardware-only One UI behavior remains for #206.

Acceptance

Assistant invocation remains usable with the ordinary UI closed, local playback stops immediately on interruption, stale server work cannot speak, and reconnect/lifecycle state is honest and bounded.

Parent: #194 Depends on: #132, #175, #196, #197 Integrates with: #133, #134, #152, #159 ## Goal Harden the shared Android voice edge for system-assistant activation, application activation and background invocation on supported Samsung lifecycle paths. ## Required behavior Android owns capture/playback/audio focus and immediate local stop; zara-server owns canonical VAD/STT/routing/LLM/tools/TTS. Support partial/final transcripts, streamed response/audio, barge-in, server cancellation, Bluetooth/headset route changes, UI close/reopen, background/foreground transitions, process recreation, authentication/reconnect and role loss. Wake-word behavior must follow #195 research and must not assume unrestricted permanent microphone access. ## Tests TDD audio-focus gain/loss, route changes, local stop budget, stale audio rejection, cancellation/replay, permission revocation, UI/process death, daemon/network loss, bounded reconnect, multiple devices/principals and redacted diagnostics. Hardware-only One UI behavior remains for #206. ## Acceptance Assistant invocation remains usable with the ordinary UI closed, local playback stops immediately on interruption, stale server work cannot speak, and reconnect/lifecycle state is honest and bounded.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
nsaspy/zara#203
No description provided.