S10 · Voice Recognition (speaker identification)
Declared state: built, not well structured.
This is the repository plan and evidence, not a live health reading. A task marked done may still depend on credentials, devices, deployment or a newer acceptance test.
Dependencies
S08
Tasks and gates
10.F1 · The gate scores the **utterance**, not the room (N2, fixed 2026-08-11).
Tier F · done · Verification: hermetic-test
Gate: `test_speaker_gate_scoping.py`
- unverified-import · undated ·
docs/SYSTEMS-DETAILS.md:1754
10.F2 · Cold-start warm-up does not mis-score the first utterance. *gate:* `test_speaker_warmup.py`
Tier F · done · Verification: hermetic-test
Gate: `test_speaker_warmup.py`
- unverified-import · undated ·
docs/SYSTEMS-DETAILS.md:1756
10.F3 · The threshold is derived from the enrolled profile's measured separation, not a constant.
Tier F · done · Verification: hermetic-test
Gate: `test_phase5_identity_bench.py` [10.F3] — threshold computed, asserted inside the
- unverified-import · undated ·
docs/SYSTEMS-DETAILS.md:1757
10.R1 · Three-way verdict — owner · not-owner · **unsure** — with unsure routed to a soft The 0.75 threshold moves up: a false "owner" executes or leaks, a false "not-owner" costs a PIN (D-15).
Tier R · open · Verification: planned-test
Gate: new `test_speaker_verdict.py` - planned gate. new `test_speaker_verdict.py`
10.R2 · Diarization for two speakers in the room: attribute each utterance.
Tier R · open · Verification: planned-test
Gate: new `test_speaker_verdict.py` - planned gate. `test_speaker_verdict.py` [two-speaker]
10.R3 · Media rejection: TV and music voices never pass the gate.
Tier R · open · Verification: planned-test
Gate: new `test_speaker_verdict.py` - planned gate. `test_speaker_verdict.py` [broadcast audio fixtures]
10.E1 · A week of real turns with zero owner false-rejects and zero stranger accepts, no manual
Tier E · open · Verification: planned-test
Gate: new `test_voice_profile_drift.py` - planned gate. recorded scores + `test_voice_profile_drift.py`
HUD screen contracts
- Voice Recognition (speaker identification) · /systems/S10
All subsystems · Source: docs/systems-plan.json.