Skip to content

[UNSELECTED ALTERNATIVE — ARCHIVED] SonKu P9.1/P9.2 early music/bass trial - #718

Closed
masarray wants to merge 26 commits into
research/sonku-p90-external-speaker-20261115from
research/sonku-p912-music-fidelity-physical-bass-20261116
Closed

masarray wants to merge 26 commits into
research/sonku-p90-external-speaker-20261115from
research/sonku-p912-music-fidelity-physical-bass-20261116

Conversation

@masarray

@masarray masarray commented Nov 15, 2026 •

Copy link
Copy Markdown
Owner

ARCHIVED / DO NOT MERGE AS THE CURRENT P9.1/P9.2 BASELINE. PR #718 and the reconciled PR #719 diverged; they are not the same PCM implementation. The user-confirmed checkpoint is issue #720, SHA 763706a15a2617a10d23b5e2e43eb5689e454ccb. Branch and commits here are kept as an alternative acoustic research record, not deleted. Current ImerSip adopts #719 through #722.


Single milestone P9.1+P9.2 — music fidelity + physical bass foundation

Reconciled baseline: exact P9.0 #717 HEAD cd85ffee, branched separately; original P8.2/P9.0 are unchanged A/B references. Current production main = 37985f02 includes #713 audio continuity. Because P9 ancestry diverged from main, this PR targets #717 and MUST NOT be merged straight to main. Approved GUI and main audio/device recovery intentionally untouched.

Native DSP opt-in mode 10 (v12, same 63-byte state):

  • Stable fixed measured KEMAR HRIR magnitude-based programme-level calibration (prepared off audio).
  • LF below ~220Hz follows the full measured 2×2 HRIR field in the same native graph, instead of returning untouched headphone L/R.
  • Three low-frequency axial room modes with computed room dimensions, source/listener world coupling, bounded Q7.5–10 resonators, no oscillator or extra audio bus; room energy goes through speaker HRIR.
  • Gentle common +1.14x 105Hz low shelf on binaural ears, no extra panner; independent speaker signals and anti-phase remain measurable.
  • Lower, separate P9.2 early/diffuse ratio rather than multiplying old P8.9 wet boosts, with room modes limited to P9.2.
  • GUI first-entry audition scene: front STATIC 82°, 1.6m, Large, Reflections 72%, Reverb 38%, Output 0dB, Mix 100%. Original plugin defaults and saved .smix never silently changed.
  • Native capability p92MusicBassFoundation, schema v12 accepted on real prepared instance only. Option is stable with P9/P8 nested flags, no re-creating option list while open.

Qualification files: new source PCM test, native JUCE v12 ACK/state/restore test, 40-readback WebView2 dropdown smoke, exact-source Linux and Windows GUI/CTEST build pipeline, and detailed docs/SONKU_P912_MUSIC_FIDELITY_PHYSICAL_BASS_FAT_20261116.md.

No fabricated dearVR parity: This is a physically motivated axial LF approximation, not a measured BRIR or cloned proprietary speaker cabinet. Full LUFS-matched listening FAT and actual headphone externalization still pending. Do not publish a release from a failed build.

…pled axial LF room modes in single native graph
…st-audition stage without rewriting sessions
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 15, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: 5da7e8d
Status: ✅  Deploy successful!
Preview URL: https://b391cb1e.arsonmix.pages.dev
Branch Preview URL: https://research-sonku-p912-music-fi.arsonmix.pages.dev

View logs

Copy link
Copy Markdown
Owner Author

P9.1+P9.2 reconciliation and measured source-audio qualification (2026-11-16)

Lineage: Research-only PR #718 -> P9.0 draft #717, NOT production main. Main 37985f02 contains #713 device continuity. P9 lineage diverged (257 ahead / 4 behind when audited); Main.cpp/CI/CMake and device recovery must be reconciled separately, not overwritten. No release or main merge authorized.

Root-cause discovery by the new failing test: Enabling the full measured low-frequency 2x2 FIR by itself caused an unacceptable energy collapse vs P9.0, mono 0.30x and anti-phase 0.14x. This was rejected. The replacement P9.2 combines measured HRIR above the existing ~220 Hz crossover with a smoothly head-relative two-source to two-ear LF speaker matrix, preserving bass while keeping its sources world-anchored. A three-mode per-room LF resonator is separately excited by each world-space virtual speaker and then passed through the existing FIR. LF matrix is a bounded psychoacoustic approximation, not measured BRIR or dearVR proprietary DSP.

Real C++ source gate, 44.1/48 kHz, preflight run 43768889621: PASS. P9 reference tests and both dropdown stability tests PASS. Test stimulus stereo normal P9.2 output energy was 1.055x (44.1k) / 1.027x (48k) P9.0, vs original dry input -0.44 dB at both sample rates at 0 dB output trim (a test-specific RMS/energy metric, NOT a music LUFS claim). Original anti-phase material remains around 0.38x P9.0 energy and must be explicitly monitored in listening FAT. Native test covers 64/128/256 callback bit parity, 0%-mix exact dry restore, source yaw, LF room ON/OFF and peak bounds.

New first-insert quality default: in P92-only JUCE processor constructor, newly inserted SonKu starts mode10/music stage (static 82° stereo/1.6m Large/reflections72%/reverb38%/gain0dB), rather than P8.2 default mode0 and -2.5dB. Saved legacy .smix states retain exact authority; no rewrite/migration. A new JUCE v12 ACK/first-insert/v1-v11 regression test is committed. The exact-head MSVC/Windows portable FAT must pass before an executable is recommended, and a loudness-matched headphone FAT against dry and dearVR remains mandatory.

Copy link
Copy Markdown
Owner Author

Exact-head Windows FAT artifact qualified (source 5da7e8d0)

GitHub Actions run 43769565823 completed successfully: Linux exact-source PCM (P9/P9.2 + dropdown) and fast source contracts PASS; native Windows MSVC configured and built real JUCE mixer EXE, P8.9/P9.0/P9.2 source+JUCE 6/6 CTests PASS, packaging verification and artifact upload PASS. The source commit was 5da7e8d0c62934bf2be519e9f3a3c66713064bc0.

Published private research portable FAT #14184142249. The 8.31MB ZIP opens without corruption and contains the verified 15,077,888-byte ArSonMix-GUI10CH-P012-Windows-x64.exe, HRTF license, SONKU_P92_FIELD_FAT.md and BUILD_PROVENANCE.json. EXE SHA-256 7c8064f37d765e83e9840b7eb13cb0e27cadabaca0e897c60cb6879cc5bcfb69 matches provenance; state schema 12, P9.1/P9.2 mode and source SHA identified correctly.

Limitations: P92 default is now actual native first-insert mode10; old .smix sessions retain their own accepted mode. Source synthetic stereo signal is -0.44 dB vs dry but not measured programme LUFS. Anti-phase P92 energy ~38% of P9 reference remains an explicit headphone FAT risk. Not a dearVR parity result; not tested on user hardware. No merge to main, no public release. The current research branch does not yet integrate production #713 audio recovery; resolving 257/4 diverged ancestry requires a separate guarded integration branch.

@masarray masarray changed the title SonKu P9.1+P9.2 (DRAFT): calibrated music fidelity + world-anchored physical bass foundation [UNSELECTED ALTERNATIVE — ARCHIVED] SonKu P9.1/P9.2 early music/bass trial Nov 16, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant