Second, deeper pass over tools/gateway/hermes_cli plus first pass over
the trees wave 1 missed (acp, acp_adapter, skills, computer_use, docker,
dashboard, conformance, monitoring, secret_sources, hermes_state,
providers). Same rubric as wave 1 (AGENTS.md test policy); security,
alternation/caching invariants, issue-number regressions, and E2E kept.
Real test-quality fixes found and rooted out along the way:
- tests/tools/test_command_guards.py made real auxiliary-LLM HTTPS calls
(DEFAULT_CONFIG smart-approval leaked in) — pinned approval
mode=manual via autouse fixture: 17.4s → 0.4s.
- test_model_switch_custom_providers.py / test_user_providers_model_switch.py
silently probed live provider catalogs (~2s/test) — stubbed
cached_provider_model_ids/provider_model_ids/fetch_api_models.
- test_telegram_noise_filter.py: 15-platform copy-paste matrix over
shared gateway.run logic → 3 representative platforms (55s → 3.9s).
- test_gateway_shutdown.py: stop()'s 5s interrupt-deadline loop spun on
MagicMock agents — interrupt.side_effect now clears _running_agents
(22s → 1.0s).
- test_gateway_inactivity_timeout.py poll-harness timings shrunk 3-5x
(24s → 1.1s); test_mcp_stability.py backoff/SIGTERM-grace sleeps
patched (15.4s → 2.5s); test_async_delegation.py negative-drain wait
5s → 0.5s.
- test_telegram_init_deadline.py: loop-block margin restored to 1.0s
with rationale comment — the watchdog-dump assertion needs the loop
blocked well past deadline+grace under parallel load (flaked once in
the 40-worker verification run at a 0.2s margin).
Verification: full hermetic suite via scripts/run_tests.sh —
2,438 files, 21,718 tests passed, 0 failed, 293.9s wall.
Suite totals vs original baseline: 46,820 → 19,757 test functions
(−57.8%), wall 583.5s → 293.9s (−50%), subprocess CPU 13,564s → 11,623s.
Two follow-ups from the voice PR (#70509).
1. macOS ARM64 onnx migration. Existing users who pinned
openwakeword.inference_framework=onnx before the tflite fix landed kept a
wake word that arms but never fires (ONNX's embedding model is broken on
Apple Silicon, upstream #336). New resolve_inference_framework() honors an
explicit framework everywhere ONNX actually works, but coerces the one
provably-dead combination (explicit onnx + macOS ARM64) to tflite with a
one-time warning. No config mutation; empty still falls back to the platform
default. Both read sites (engine init + requirements check) route through
the shared resolver.
2. Voice turn-timeout leak. Each listen cycle reassigned turnTimeoutRef
without clearing the prior 60s timer, so a stale timer from an earlier
cycle could fire handleTurn() mid-way through a later listen — after enough
idle re-listens this wedged the loop into a non-re-arming state (the
'voice chat deactivates after ~a minute' report). Clear before re-arm.
Tests: 64 wake tests (added onnx-coercion / intel-kept / tflite-kept /
empty-default cases; updated the stale 'explicit onnx kept on ARM64' test
that encoded the old broken behavior), 39 desktop voice/wake vitest, tsc +
eslint clean.
'hey hor' triggered the wake word: the default sensitivity was 0.5, which
for openWakeWord IS the raw per-frame score threshold — openWakeWord's own
permissive baseline that near-misses clear. Raised the default to 0.6 so
phonetic near-misses fall short while real 'hey hermes' (typically 0.9+)
still fires easily.
Also fixed a real cross-engine inconsistency found while checking: the
sensitivity knob is documented 'higher = stricter' and behaves that way
for openWakeWord (threshold = sensitivity) and sherpa (0.05 + 0.4*s), but
Porcupine's own 'sensitivities' param runs the opposite way (higher = MORE
false alarms, per Picovoice). Turning sensitivity up made Porcupine looser
— backwards. Now inverted (1 - sensitivity) so 'higher = stricter' holds
for every engine.
- tools/wake_word.py: default 0.6; _sensitivity fallback uses _DEFAULTS;
Porcupine sensitivity inverted with rationale
- hermes_cli/config.py + docs: default + consistent-direction note
- tests: Porcupine inversion, default>=0.6 regression, fallback-to-default
openWakeWord scores one ~80ms frame at a time and the detector fired the
instant a SINGLE frame crossed threshold — so a stray phoneme in background
conversation could trigger the wake word unintentionally (reported in
testing). A real utterance of the phrase holds a high score across several
consecutive frames; an ambient blip spikes just one.
_OpenWakeWordEngine now requires N consecutive over-threshold frames
(wake_word.confirmation_frames, default 3) before firing. The streak resets
on any sub-threshold frame and on engine reset() (pause/resume), so a
pre-pause frame can't count toward a post-resume fire. confirmation_frames=1
restores the old single-frame behavior; clamped 1..10.
Only openWakeWord is affected — sherpa (streaming transducer) and porcupine
decode the whole phrase internally and already reject single-frame spikes.
- tools/wake_word.py: _confirmation_frames() accessor, streak logic in
process()/reset(), config default
- hermes_cli/config.py: wake_word.confirmation_frames documented default
- tests: 5 new (spike rejected, sustained fires once, =1 legacy behavior,
reset clears streak, config clamp) — 58 wake tests green
- docs: 'Reducing false triggers on ambient speech' section
Follow-up to benbarclay's macOS ARM64 tflite fix:
- test_openwakeword_bundled_model_matches_framework stubs
ensure_tflite_runtime()=True so it exercises artifact selection, not
runtime availability — off-Darwin the bridge legitimately returns
False and the engine falls back to onnx (that path is covered by its
own tests). Was failing on Linux CI otherwise.
- check_wake_word_requirements now ANDs both this branch's STT/TTS gate
and Ben's tflite_ok into 'available' (cherry-pick conflict resolution).
openWakeWord's ONNX backend returns near-zero scores on Apple Silicon
(dscripka/openWakeWord#336), so "Hey Hermes" never crossed the 0.5
threshold: the listener armed, the microphone worked, and nothing fired.
Bisecting the pipeline puts the fault in exactly one stage — feeding the
same audio through both backends, the melspectrogram front-end is
bit-identical (maxdiff 0.00000) and the wake classifier agrees on
identical features, while the shared embedding model diverges by 45.44.
Cross-feeding confirms it: tflite features scored through the *onnx*
classifier give 0.9948 vs 0.000009 for onnx features. A telling
secondary symptom is that scores fall as input gets louder (0.5x ->
0.00031, 8x -> 0.000066), which is garbage inference rather than a weak
detection.
Selecting tflite in config alone does not fix it. openWakeWord hardcodes
`import tflite_runtime.interpreter` but declares tflite-runtime for
`platform_system == "Linux"` only; on macOS the equivalent wheel is
ai-edge-litert, so that import always fails and model.py silently
downgrades back to onnx. The result is a detector that reports itself
listening and can never fire.
- default the backend per platform (tflite on macOS ARM64, onnx
elsewhere) instead of hardcoding onnx, and pick the matching bundled
model artifact
- bridge tflite_runtime -> ai_edge_litert through sys.modules, in-process,
with no writes to site-packages
- refuse the silent onnx downgrade on macOS ARM64 and report the missing
runtime through check_wake_word_requirements() so the GUI surfaces an
actionable hint rather than arming a dead ear
- lazy-install ai-edge-litert via its own feature key, because lazy-dep
specs cannot carry PEP 508 markers (_spec_is_safe rejects ";")
An explicit `inference_framework` in config still wins, so anyone pinning
a backend keeps it.
Verified on macOS 26.5.2 / M-series: "hey hermes" scores 0.0005 on onnx
and 0.9423 on tflite from the same clip, with cross-phrase controls at
0.0003. Live over-the-air through the real microphone fires 4/4
utterances (peak 0.9532).
Two fixes from live testing (Teknium):
1. Desktop installs now ship the wake/voice stacks up front.
install.sh + install.ps1 desktop stages run 'uv pip install
-e .[wake,voice]' (best-effort, lazy-install remains the fallback)
before building the app, so the first ear-click arms instantly
instead of sitting through a multi-minute onnxruntime download.
CLI-only installs keep the lazy path — [all] curation unchanged.
2. The vanished ear: the STT/TTS gate made wake.status call
check_tts_requirements(), whose edge path runs _import_edge_tts →
lazy_deps.ensure — a synchronous PIP INSTALL inside a status poll.
On a venv without edge-tts that blew the desktop's 30s RPC timeout,
armWakeWord caught the error, the atom never learned enabled=true,
and the ear unmounted. _tts_ready is now a pure probe: deps missing
+ lazy installs allowed counts as ready (installs at first speak)
WITHOUT touching pip; check_tts_requirements only runs once deps
are present. Regression test asserts the probe never calls it while
deps are missing.
The wake loop is wake → record → STT → agent → TTS. Arming without
either end configured gives a mic that hears you and then does nothing
perceivable — a useless experience. check_wake_word_requirements now
probes both (same probes /voice uses: stt.enabled + provider != none;
check_tts_requirements) and refuses with a pointer to `hermes tools`
naming exactly which half is missing. The desktop ear hides (available:
false already hides the button), /wake on prints the hint on CLI/TUI.
wake.start also validates requirements BEFORE persisting
wake_word.enabled, so a refused gesture can't leave config claiming on
while nothing can ever arm.
Tests: per-half and both-missing hint assertions; existing requirements
tests pinned via _voice_loop_ready so they don't depend on the test
venv's installed voice stack. E2E: stt.enabled=false in a real config
-> unavailable with the speech-to-text hint.
Internal testing (macOS): clicking the ear froze the button for ~30s,
then it went blue but 'hey hermes' never fired even though STT worked.
Two distinct bugs:
1. Frozen-then-timeout button: first-use wake.start lazy-installs the
detection engine (onnxruntime is a large wheel), which blows the
desktop's default 30s WS request timeout — the RPC 'failed' client-
side while the backend kept installing and armed later on its own.
wake.start now gets a 180s budget and the pending state says
'arming — first use may take a minute while the engine installs'
instead of a silent disabled button.
2. Armed but deaf: macOS grants mic permission per PROCESS. The
renderer having mic access (STT working) does not grant the Python
backend anything — CoreAudio hands an unentitled process a 'working'
stream that delivers zeros forever, so the listener looks healthy
and can never hear the phrase. The detector now tracks frame peaks:
10s of consecutive near-zero frames sets audio_silent, logged with
the exact macOS Settings path, cleared automatically when audio
appears. Surfaced everywhere: wake.status (audio_silent + hint),
desktop ear tooltip (kept visible while listening), /wake status on
TUI (⚠ line) and classic CLI, plus a docs troubleshooting section.
Tests: detector silent-flag set/recover cycle (fake silent/loud
streams), desktop tooltip keeps the dead-mic hint while listening.
44 wake Python tests, 22 desktop + 14 TUI vitest green.
check_wake_word_requirements() gated 'available' on the audio probe,
but the probe imports sounddevice + numpy — two of the packages the
lazy installer would install. On a fresh machine deps_ok was False, so
audio_ok was always False and /wake on printed the manual pip hint and
bailed before the engine constructors' lazy_deps.ensure() could run.
Now the audio probe only runs once deps are installed; with deps
missing and lazy installs allowed (the default), /wake on proceeds and
ensure() installs the pinned engine deps in-process — no restart. The
manual pip hint remains for security.allow_lazy_installs=false, and a
mic hint still blocks when deps are present but no audio device works.
The CLI announces the one-time engine install so the pause is explained.
numpy is a lazy voice-extra dep absent from CI's hermetic env; the two
engine process() tests imported it for real. Stub asarray/float32 in
the fixture (verified against a blocked-numpy import, matching CI).
One sherpa listener now enrolls every wake-enabled profile's phrase
(defaulting to "hey <profile name>") and reports WHICH phrase fired.
wake.detected gains a profile field; the desktop live-switches to the
matching profile (same path as the profile rail), opens a fresh session
there, and starts hands-free voice. The single-profile CLI/TUI print the
hermes -p switch command for foreign-profile phrases instead of
answering as the wrong profile. Opt out per listener with
wake_word.profile_routing: false.
New "sherpa" wake_word provider: the configured phrase is BPE-tokenized
at runtime against a small streaming zipformer KWS model (~13 MB English,
one-time download cached under HERMES_HOME), so ANY typed phrase works
with zero training — including per-profile phrases like "hey coder".
wake.sherpa lazy-dep group + [wake] extra grow sherpa-onnx/sentencepiece;
requirements probe routes per provider; sensitivity maps onto sherpa
keywords_threshold. E2E-verified on real audio (target phrase fires,
foreign phrase stays silent, reset drops buffered state).
From #53378: ships hey_hermes.onnx/.tflite (openWakeWord pipeline,
Apache-2.0) under tools/wakewords/, resolves the default (and hey_hermes
aliases) to the bundled file, ensures openWakeWord base feature models
are fetched for custom paths too, and updates config defaults + docs
from hey_jarvis to hey hermes.
Two bugs surfaced by the desktop wake conversation:
1. Runaway loop: wake -> voice -> resume -> wake fired again within
~200ms. openWakeWord keeps its rolling feature buffer across
pause/resume, so on resume it immediately re-scored the "hey jarvis"
captured before the pause and re-fired, reopening a session and
restarting voice in a tight cycle. Reset the engine buffer on every
detector (re)start so resume begins from clean audio.
2. Empty-transcript toast: a silent re-listen returns
success:false / "… STT returned empty transcript", which the desktop
transcribe endpoint turned into a 400 -> thrown error -> "Voice
transcription failed" notification on every silent gap. Treat an empty
transcript as no-speech: return {ok, transcript: ""} so the voice loop
quietly re-listens. Real failures still 4xx/5xx.
Makes the wake word a tri-surface feature with one configurable owner.
- wake_word.surface ("auto" | "cli" | "tui" | "gui") + shared
wake_surface_enabled() gate consulted by every surface, so exactly one
place owns the listener and the new session it opens.
- tui_gateway: wake.start/stop/pause/resume/status RPCs + a wake.detected
event, sharing one server-side detector for both TUI and desktop. The
detector yields the mic to voice.record (pause on capture start, resume
on terminal) and to the desktop's browser mic (wake.pause/resume).
- TUI (Ink): arm wake.start on gateway.ready; on wake.detected open a
fresh session and start voice capture.
- Desktop (Electron): arm wake.start on connect; on wake.detected open a
fresh session.
- CLI now gates on wake_surface_enabled("cli"); /wake status shows surface.
- Tests for the surface gate; docs cover the surface knob + cross-surface.
Adds an opt-in, on-device hotword listener for the CLI. With
wake_word.enabled (or /wake on), Hermes listens in the background for a
wake phrase; on detection it starts a fresh session, captures one
utterance through the existing voice pipeline, and answers — the
"Hey Siri" pattern.
- tools/wake_word.py: provider-pluggable detector (openWakeWord, free
local default; Porcupine, premium) over the shared 16 kHz sounddevice
capture path. Background daemon thread with pause/resume so it yields
the mic during a voice turn.
- CLI wiring: startup listener (off-thread), on-wake flow, an idle
watchdog that resumes the detector after each turn, cleanup hook, and
a /wake [on|off|status] command.
- config.yaml wake_word section; PORCUPINE_ACCESS_KEY as an optional
secret. Engines lazy-install via the [wake] extra.
- Hands a transcript to the input queue exactly like voice mode, so no
system-prompt/cache mutation. No new core model tool.
- Tests (mocked, no live audio/network) + feature docs.
