Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
26 commits
Select commit Hold shift + click to select a range
97de133
test(native): prove Windows EXE embeds approved Preview GUI byte-for-…
masarray Nov 14, 2026
4809937
feat(mixer): reconcile native global FX A/B, GAIN/WIRE and centered P…
masarray Nov 15, 2026
96ad88b
fix(test): eliminate SonKu Win32/Windows CTest fixture stack overflow…
masarray Nov 15, 2026
37985f0
fix(audio): restore approved device and MAIN after disconnect (#713)
masarray Nov 15, 2026
ea0a7c0
merge(sonku): reconcile P9 acoustic tree with main P0 audio continuit…
masarray Nov 15, 2026
25d737a
feat(sonku): fixed-size phase-complemented world-speaker LF and bound…
masarray Nov 15, 2026
d2f8488
feat(sonku): P9 stereo full-frequency loudness foundation and 2-sourc…
masarray Nov 15, 2026
08124b9
test(sonku): real P9 PCM LF room, world-yaw, phase, anti-phase, 44.1/…
masarray Nov 15, 2026
9a0f06b
build(sonku): compile P9.1+P9.2 real source PCM regression under same…
masarray Nov 15, 2026
8ef2621
fix(sonku): prohibit P8 headphone-locked LF restoration overriding P9…
masarray Nov 15, 2026
5e33c1f
feat(sonku): first P9 factory music-stage scene 82°/near/room without…
masarray Nov 15, 2026
ff85f79
ci(sonku): qualify reconciled P9.1/P9.2 world-bass native PCM, JUCE s…
masarray Nov 15, 2026
55d939f
fix(sonku): eliminate double-counted P8 bass shelf and align new P9 w…
masarray Nov 15, 2026
73f1509
fix(sonku): eliminate double-counted P8 bass shelf and align new P9 w…
masarray Nov 15, 2026
c06be61
test(sonku): enforce deterministic source-referenced P9 LF loudness c…
masarray Nov 15, 2026
c017440
test(sonku): log measured P9/source level ratio before strict fidelit…
masarray Nov 15, 2026
24e4962
fix(sonku): source-referenced LF normalization after removing old dou…
masarray Nov 15, 2026
8a00db6
test(sonku): isolate and prove genuinely world-rotating 70 Hz LF with…
masarray Nov 15, 2026
d648cb3
docs(sonku): explain main reconciliation, source-referenced music lev…
masarray Nov 15, 2026
ff705ac
ci(sonku): bundle exact P9.1/P9.2 LF FAT and reconciliation evidence …
masarray Nov 15, 2026
54cd380
fix(sonku): reset only bounded LF/coherence state across P9 native A/…
masarray Nov 15, 2026
475ae71
fix(sonku): advertise reconciled P9.1/P9.2 capability only from prepa…
masarray Nov 15, 2026
3a81e86
fix(sonku): show P9.1+P9.2 Music + Physical Bass only with matching n…
masarray Nov 15, 2026
c663535
test(sonku): prevent P9.0 mislabel, assert P9.1+P9.2 native gated sel…
masarray Nov 15, 2026
1f2f5d2
ui(sonku): keep full P9.1 + P9.2 mode name readable in compact native…
masarray Nov 15, 2026
763706a
test(sonku): assert compact fully visible P9.1+P9.2 selector caption
masarray Nov 15, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 21 additions & 5 deletions .github/workflows/p1-windows.yml
Original file line number Diff line number Diff line change
Expand Up @@ -180,6 +180,18 @@ jobs:
node --check web/approved-native-bridge.js
node --check web/approved-native-contract.js
node --test --test-name-pattern="ASIO4ALL|ASIO restart" tests/mixer_io_operator_contract_test.mjs
node --test tests/audio_continuity_contract_test.mjs
# This is also in the full Windows contract set; enforce it on the
# fast runner so a stale device-persistence assertion cannot waste
# a complete JUCE/MSVC job again.
node --test tests/direct_restore_recovery_contract_test.mjs
g++ -std=c++20 -O2 -Wall -Wextra -Werror tests/audio_continuity_policy_test.cpp -o "$RUNNER_TEMP/armix-audio-continuity"
"$RUNNER_TEMP/armix-audio-continuity"
# SonKu has >192 KiB acoustic rings per DSP instance. Prove the
# actual PCM assertions under a Windows-sized 1 MiB stack *before*
# allocating a full MSVC runner; never skip the native Windows CTest.
g++ -std=c++20 -O2 -Wall -Wextra -Werror tests/sonku_spatial_core_test.cpp -o "$RUNNER_TEMP/armix-sonku-spatial-core"
(ulimit -s 1024; "$RUNNER_TEMP/armix-sonku-spatial-core")
g++ -std=c++20 -O2 -Wall -Wextra -Werror tests/first_run_endpoint_policy_test.cpp -o "$RUNNER_TEMP/armix-first-run-policy"
"$RUNNER_TEMP/armix-first-run-policy"
node --test tests/native_first_run_bootstrap_contract.test.mjs
Expand All @@ -188,6 +200,7 @@ jobs:
# inspector Python child must explicitly select UTF-8.
PYTHONIOENCODING=cp1252 node --test --test-name-pattern="WIRE inspector controls PRIMARY BUS" tests/approved_native_m2_contract.test.cjs
python -m unittest discover -s tests -p approved_native_gui_bundle_test.py -v
python -m unittest discover -s tests -p test_embedded_gui_resources.py -v
python -m unittest discover -s scripts/tests -p test_factoryfx_release_bundle.py -v
# Portable ChannelId PCM routing MUST be checked in under a minute,
# before allocating a Windows JUCE/MSVC runner (93-CTest suite).
Expand Down Expand Up @@ -647,7 +660,7 @@ jobs:
sonku-p90-source:
# Runs only on the P9 research PR, even while its production jobs skip.
name: SonKu P9 exact-source C++ PCM preflight
if: ${{ github.event_name == 'pull_request' && startsWith(github.event.pull_request.head.ref, 'research/sonku-p90-') }}
if: ${{ github.event_name == 'pull_request' && (startsWith(github.event.pull_request.head.ref, 'research/sonku-p90-') || startsWith(github.event.pull_request.head.ref, 'research/sonku-p91-p92-')) }}
runs-on: ubuntu-latest
timeout-minutes: 10
steps:
Expand All @@ -664,14 +677,16 @@ jobs:
FLAGS="-std=c++20 -O2 -Wall -Wextra -Werror -pedantic -DSONKU_P7_MEASURED_ELEVATION_RESEARCH=1 -DSONKU_P71_DIRECT_XYZ_RESEARCH=1 -DSONKU_P72_NATIVE_XYZ_RT=1 -DSONKU_P75_MUSIC_SAFE_RESEARCH=1 -DSONKU_P80_TRANSPARENT_BASS_RESEARCH=1 -DSONKU_P82_MULTIBAND_ROOM_RESEARCH=1 -DSONKU_P83_PHYSICAL_SPEAKER_RESEARCH=1 -DSONKU_P84_SINGLE_CORE_FUSION_RESEARCH=1 -DSONKU_P85_WORLD_COHERENCE_RESEARCH=1 -DSONKU_P86_HORIZONTAL_FOCUS_RESEARCH=1 -DSONKU_P87_MUSIC_BODY_RESEARCH=1 -DSONKU_P88_PERCEPTUAL_ALIGNMENT_RESEARCH=1 -DSONKU_P89_DEPTH_COHERENCE_RESEARCH=1 -DSONKU_P90_TRUE_EXTERNAL_RESEARCH=1"
g++ $FLAGS tests/sonku_p90_true_external_source_test.cpp -o /tmp/sonku-p90
/tmp/sonku-p90
g++ $FLAGS tests/sonku_p91_p92_music_bass_foundation_test.cpp -o /tmp/sonku-p91p92
/tmp/sonku-p91p92
g++ $FLAGS tests/sonku_p89_stereo_room_depth_source_test.cpp -o /tmp/sonku-p89
/tmp/sonku-p89
echo "SONKU_P90_NATIVE_SOURCE_CTEST_PASSED"

sonku-p90-windows-fat:
name: SonKu P9 actual JUCE / GUI portable x64
needs: sonku-p90-source
if: ${{ github.event_name == 'pull_request' && startsWith(github.event.pull_request.head.ref, 'research/sonku-p90-') }}
if: ${{ github.event_name == 'pull_request' && (startsWith(github.event.pull_request.head.ref, 'research/sonku-p90-') || startsWith(github.event.pull_request.head.ref, 'research/sonku-p91-p92-')) }}
runs-on: windows-2022
timeout-minutes: 55
steps:
Expand Down Expand Up @@ -717,12 +732,12 @@ jobs:
- name: Build real native JUCE PCM, state ACK, and Windows GUI
shell: pwsh
run: |
cmake --build build --config Release --parallel 2 --target ARMixSonKuP90NativeTests ARMixSonKuP90SourceTests ARMixSonKuP89NativeTests ARMixSonKuP89SourceTests ARMixConsole
cmake --build build --config Release --parallel 2 --target ARMixSonKuP91P92MusicBassTests ARMixSonKuP90NativeTests ARMixSonKuP90SourceTests ARMixSonKuP89NativeTests ARMixSonKuP89SourceTests ARMixConsole
if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
if (-not (Test-Path -LiteralPath 'build/ARMixConsole_artefacts/Release/AR MIX Studio.exe' -PathType Leaf)) {
throw 'SONKU_P90_NATIVE_GUI_EXE_MISSING'
}
ctest --test-dir build -C Release --output-on-failure -R '^NativeSonKuP(89RoomAnchoredStereoAndP88GoldenParity|89JuceV10AppliedAckAndV9Restore|90StereoExternalizationSource|90JuceV11AppliedAckAndLegacyRestore)$'
ctest --test-dir build -C Release --output-on-failure -R '^NativeSonKuP(89RoomAnchoredStereoAndP88GoldenParity|89JuceV10AppliedAckAndV9Restore|90StereoExternalizationSource|90JuceV11AppliedAckAndLegacyRestore|91P92MusicFidelityWorldBassAndRoom)$'
if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
- name: Verified portable package, no installer or release
shell: pwsh
Expand All @@ -735,12 +750,13 @@ jobs:
if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
Copy-Item docs/THIRD_PARTY_8D_SONKU_HRTF.md dist/THIRD_PARTY_8D_SONKU_HRTF.md
Copy-Item docs/SONKU_P90_TRUE_EXTERNAL_SPEAKER_FAT_20261115.md dist/SONKU_P90_FIELD_FAT.md
Copy-Item docs/SONKU_P91_P92_RECONCILED_MUSIC_FIDELITY_BASS_FAT_20261116.md dist/SONKU_P91_P92_FIELD_FAT.md
$provenance = Get-Content -LiteralPath dist/BUILD_PROVENANCE.json -Raw | ConvertFrom-Json
if ($provenance.sourceCommit -ne (git rev-parse HEAD).Trim()) {
throw 'SONKU_P90_SOURCE_COMMIT_MISMATCH'
}
$provenance | Add-Member -Force -NotePropertyName sonkuExperiment -NotePropertyValue @{
mode = 'P9.0_TRUE_EXTERNAL_SPEAKER'
mode = 'P9.1_P9.2_MUSIC_FIDELITY_PHYSICAL_BASS'
stateSchema = 11
headphoneFat = 'PENDING'
publicRelease = $false
Expand Down
33 changes: 33 additions & 0 deletions .github/workflows/ui-smoke.yml
Original file line number Diff line number Diff line change
Expand Up @@ -394,6 +394,37 @@ jobs:
if-no-files-found: ignore
retention-days: 7


mixer-global-fx-visual:
name: Global FX hardware power + PAN real Chromium (3 viewports)
needs: quick
# Deliberately scoped: one current GUI milestone, no generic full-suite,
# no unrelated SonKu, GAIN or paid Windows native rebuild.
if: ${{ github.event_name == 'pull_request' && startsWith(github.head_ref, 'feat/mixer-global-fx-ab-') }}
runs-on: ubuntu-latest
timeout-minutes: 12
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with:
python-version: "3.12"
- name: Install one Chromium
run: |
python -m pip install playwright
python -m playwright install --with-deps chromium
- name: Assemble exact PR head /preview/ assets
run: python scripts/build_pages.py --out dist-pages --sha "${{ github.event.pull_request.head.sha }}" --branch main
- name: Verify FX master power face and individual state; PAN no-collision
run: python tests/pages_global_fx_ab_pan_wire_smoke.py
- name: Retain ON/OFF screenshots from all viewports
if: ${{ always() }}
uses: actions/upload-artifact@v4
with:
name: ArSonMix-Global-FX-PAN-${{ github.run_id }}
path: arsonmix-global-fx-*.png
if-no-files-found: ignore
retention-days: 7

frontend:
name: Full Chromium /preview/ visual and interaction qualification
# Draft edits are cheap; ChatGPT milestone comments dispatch this full suite.
Expand Down Expand Up @@ -470,6 +501,8 @@ jobs:
run: python tests/approved_preview_mobile_navigation_smoke.py
- name: FX rack master bypass preserves each insert state (real Chromium)
run: python tests/approved_fx_rack_master_power_smoke.py
- name: Global FX A/B, PAN extremes and GAIN/WIRE state (real Chromium)
run: python tests/pages_global_fx_ab_pan_wire_smoke.py
- name: Native FX pointer transfer, delete and editor identity
run: python tests/approved_native_fx_edit_smoke.py
- name: Native EQ real Chromium SVG gestures, ACK, graph and responsive editor
Expand Down
24 changes: 24 additions & 0 deletions CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -338,6 +338,10 @@ target_link_libraries(ARMixConsole PRIVATE

include(CTest)
if(BUILD_TESTING)
# Pure-C++ P0 audio-continuity gate: display idle must never reset MAIN.
add_executable(ARMixAudioContinuityPolicyTests tests/audio_continuity_policy_test.cpp)
target_compile_features(ARMixAudioContinuityPolicyTests PRIVATE cxx_std_20)
add_test(NAME P0DisplayIdlePreservesAudioAndDeviceRecoveryIntent COMMAND ARMixAudioContinuityPolicyTests)
juce_add_console_app(ARMixFactoryFxRegistryTests PRODUCT_NAME "ArSonMix Factory FX Integrity Tests")
target_sources(ARMixFactoryFxRegistryTests PRIVATE tests/factory_fx_registry_test.cpp src/FactoryFxRegistry.h)
target_link_libraries(ARMixFactoryFxRegistryTests PRIVATE juce::juce_core juce::juce_cryptography juce::juce_recommended_config_flags)
Expand Down Expand Up @@ -793,6 +797,26 @@ if(BUILD_TESTING)
SONKU_P90_TRUE_EXTERNAL_RESEARCH=1)
add_test(NAME NativeSonKuP90StereoExternalizationSource
COMMAND ARMixSonKuP90SourceTests)
add_executable(ARMixSonKuP91P92MusicBassTests
tests/sonku_p91_p92_music_bass_foundation_test.cpp)
target_compile_features(ARMixSonKuP91P92MusicBassTests PRIVATE cxx_std_20)
target_compile_definitions(ARMixSonKuP91P92MusicBassTests PRIVATE
SONKU_P7_MEASURED_ELEVATION_RESEARCH=1
SONKU_P71_DIRECT_XYZ_RESEARCH=1
SONKU_P72_NATIVE_XYZ_RT=1
SONKU_P75_MUSIC_SAFE_RESEARCH=1
SONKU_P80_TRANSPARENT_BASS_RESEARCH=1
SONKU_P82_MULTIBAND_ROOM_RESEARCH=1
SONKU_P83_PHYSICAL_SPEAKER_RESEARCH=1
SONKU_P84_SINGLE_CORE_FUSION_RESEARCH=1
SONKU_P85_WORLD_COHERENCE_RESEARCH=1
SONKU_P86_HORIZONTAL_FOCUS_RESEARCH=1
SONKU_P87_MUSIC_BODY_RESEARCH=1
SONKU_P88_PERCEPTUAL_ALIGNMENT_RESEARCH=1
SONKU_P89_DEPTH_COHERENCE_RESEARCH=1
SONKU_P90_TRUE_EXTERNAL_RESEARCH=1)
add_test(NAME NativeSonKuP91P92MusicFidelityWorldBassAndRoom
COMMAND ARMixSonKuP91P92MusicBassTests)
juce_add_console_app(ARMixSonKuP90NativeTests
PRODUCT_NAME "ArSonMix SonKu P90 JUCE v11 State Tests")
target_sources(ARMixSonKuP90NativeTests PRIVATE
Expand Down
47 changes: 47 additions & 0 deletions docs/SONKU_P91_P92_RECONCILED_MUSIC_FIDELITY_BASS_FAT_20261116.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
# SonKu P9.1 + P9.2 — Reconciled Music Fidelity and World-Anchored Bass

**Status:** PRIVATE HEADPHONE FAT CANDIDATE — NOT RELEASED, NOT dearVR PARITY

## Reconciliation (performed first)

- Starting P9 HEAD `cd85ffee054dad6b3b752b5351d34921ddc92717` (draft PR #717).
- Production `main` `37985f02caecd93b22f853183888d81e18dc6736` already contains #713 ASIO/WASAPI continuity, operator MAIN intent, audio graph/global FX and saved-session safeguards.
- New two-parent reconciliation commit `ea0a7c0a5bb3afb2c4295176447c2eb78bb6835b`, with main as direct ancestor. Imported the 30 main-changed files and applied 29 exact-context patch hunks for the four overlapping files: `src/Main.cpp`, `CMakeLists.txt`, `p1-windows.yml`, `mixer-fx-rack.css`. No production branch changed or merged.
- The new acoustic development and GUI belong to `research/sonku-p91-p92-fidelity-bass-reconciled-20261116`, PR #719, based on PR #717.

## Why the previous default was not acceptable

Old P9 used legacy output trim 0.75 (about -2.5 dB) and HRTF wet factor 0.5. Critically, P90 inherited P8.2 `transparentBass`: the low part of the world HRTF was *subtracted* and the original headphone-locked L/R bass was brought back. That violated the world-speaker premise. The original P8.2 bass-shelf 0.82 (+5.2 dB when fully active) was also still applied to the P9 branch with room ON, so a naive LF enhancement stacked gains and clipped musical intent.

The operator's field setting (82° speakers, near 1.6m, Large 28.8m room, +7 dB output, reflections 100%, reverb 62%) is a crucial listening reference; copying that full +7 dB and 100% wet as default would hide the bug.

## Integrated native DSP architecture

1. Keep the existing measured 2x2 KEMAR direct HRIR and P9 coherence-safe MID/SIDE compositor. Only P9 mode 9 routes into the new LF foundation.
2. Complementary two-pole lowpass at 165 Hz: `HRTF(high)=HRTF - LP(HRTF)` and `worldLow = W_L*LP(L) + W_R*LP(R)`. BOTH use the same lowpass topology, avoiding a disconnected high-pass and duplicated low energy. Source weights are `0.5*(1 +/- 0.27*sin(headAngle))`, with a 361-point sine lookup prepared once. They follow the existing world-speaker and head-yaw angles; no original headphone L/R dry-low return.
3. P9 only, use 0.72 scene binaural gain and no legacy room loudness multiplier 1.12 or legacy +0.82 LF shelf. The full original peak protection remains. Native P8.2–P8.9 behavior is unchanged.
4. Room EARLY now receives independent L/R with the actual P9 speaker angle. The eight-line late FDN receives the true L/R, not a duplicate MID source. Speaker azimuth and world/head rotation apply to the early image reflections.
5. Three lightweight axial mode resonators per room (Studio 7.2, Chamber 14.4, Large 28.8m) are computed from `f=n*343/(2*L)` nearest preselected LF clusters and a stable Q=2.0/2.8/3.8 bandpass. Bounded wet weights 0.14/0.105/0.075 and 60ms room/size smoothing; ROOM OFF is strictly zero. This is *not* a personalized/physical subwoofer driver/cabinet IR, nor measured modal BRIR.
6. On the first intentional switch from an *untouched legacy factory scene* to P9, an isolated native-ACKed close music stage initializes: 82° source spread, 1.6m, Large, 78% ER, 33% FDN, 0 dB output, orbit OFF. Existing .smix and manually adjusted states do not reset. No new state version or GUI visual redesign.

## Objective PCM admission

Tests use real native C++ at both 44.1 and 48 kHz. Check full source-referenced 70/116Hz+5.2kHz steady-state input/output RMS at output gain 0dB, loudness floor/ceiling, finite peaks below the same 0.98 linked guard, 128/256 block bit parity, left/right/anti-phase, 0% ROOM modal bit-exact zero, nonzero room response, and head-yaw changes. An additional isolated *70Hz-only* WorldBass test verifies left/right ear dominance reverses when the listener rotates, so treble cannot camouflage headlocked bass.

Reference PCM result after gain correction (source sum RMS 0.40890):
- 44.1k P9 sum RMS 0.41656, approximately +0.16 dB; P8.9 0.37159.
- 48k P9 sum RMS 0.41477, approximately +0.12 dB; P8.9 0.37152.

These are NOT music integrated LUFS, not physical headphone acoustic dB, and not proof of perceptual externalization. The new isolated 70Hz rotation test and Windows JUCE CTest must pass for its own exact commit, not inherited from an earlier source run.

## Required headphone FAT

- Compare raw stereo versus P8.2, P8.9, reconciled P9.1/P9.2 and dearVR PRO 2 under matched output LUFS within +/-0.5 LU. Disable all other FX and parallel dearVR insert while testing each.
- Material: percussion kick, bass guitar/sub 30–120Hz, orchestral ensemble, centered vocal, sibilance/transients, anti-phase stereo torture and real recordings with substantial 60–180Hz.
- Positions: 82°/120°/180° speaker spread, 1.6/3/6.7m, yaw -90/0/90/180. Toggle Large/Chamber/OFF with same level; distinguish bass *impact* from low-frequency bass *boom* and transient masking.
- Analysis: frequency response, LF group delay and modal decay, true IACC/ILD/ITD, DRR, headphone correction, CPU P95 per callback, silence/denormal/NaN and resumed device handling. Existing +7dB user A/B can be monitored but must NOT become default gain.
- Record scene, capture screenshots, compare with dearVR by listening. Do not claim measured dearVR feature parity or commercial release before FAT.

## Engineering constraints

No heap, locks, SIMD prerequisites, FFT allocation, filesystem calls or extra callback inside PCM. Coefficients are prepared once, weights small fixed arrays, single native route. Room modes are low-level bounded returns, not uncontrolled feedback. Preserve previous saved session schemas v1..v11, global FX gate, MAIN continuity and ASIO priority. Production main and P8.2–P8.9 remain unmodified.
Loading
Loading