Skip to content

SonKu P9.7 DRAFT — concert bass impact, external center presence & early air - #728

Closed
masarray wants to merge 6 commits into
feat/sonku-p95-p96-distance-reinforcement-20261116from
feat/sonku-p95-p96-p97-stage-20261116
Closed

masarray wants to merge 6 commits into
feat/sonku-p95-p96-distance-reinforcement-20261116from
feat/sonku-p95-p96-p97-stage-20261116

Conversation

@masarray

Copy link
Copy Markdown
Owner

SonKu P9.7 — powerful stage punch, center presence and early-reflection air (PRIVATE DRAFT)

Selected checkpoint: P9.6 PR #727 at 720cec619b41aa601414d4b5ae664cb418f04ca1. User listening feedback: P9.6 already reasonably good, do not change the known-good sound. This PR is exactly one independent child commit, not a modification or merge of P9.6. No changes to main or old sessions.

Original user goal

Greater live-sound-system bass impact and resonance; more external/prominent center-panned vocals; treble that opens into the room; better stereo separation and a vivid externalized soundstage, WITHOUT thinning the music, mono collapse, artificial fake width, or limiter pumping.

Research-led, bounded single native DSP

  • Externalization literature: direct pinna details and binaural reflection interaural coherence/level cues are critical; raw reverb amount or stereo width does not guarantee front externalization.
  • AES virtual bass: unmanaged harmonic synthesis risks audible artifacts; this candidate deliberately does not invent subharmonics/harmonics.
  • New SonKuP97StageImpact.h, single core, P9 mode9 only, ROOM ON and distance >2m. 45ms coefficient-free per-sample depth blend, 2–6m ramp; no second engine/bus, no artificial Haas, source separation model, FFT or allocations.
  • Bass: prepared stable low-Q ~85Hz bandpass and 62Hz one-pole modest direct pressure reinforcement after existing P96 and measured KEMAR ear graph. Existing P9.2 damped world room modes get only bounded up-to-12% level (not added feedback/RT60).
  • Center presence: low-Q 1.9kHz of the original stereo MID gently reinforces BOTH virtual loudspeaker feeds before existing measured HRIR. Original stereo still drives room and world LF. This is not independent vocal extraction and affects other center-panned instruments.
  • Treble external air: brighten the high-frequency component above ~4.5kHz in only the existing real directional, per-ear early-reflection return. No additional channel crossing, headlocked copy, stereo delay or tail time change.
  • Flag ARMIX_SONKU_EXPERIMENTAL_P97_STAGE_IMPACT=OFF by default and requires opt-in P9.6; preserved native state schema v11, GUI controls, bypass, independent readback capability p97StageImpact, limiter and P9.5 gain/peak protections. New normal GUI mode labels remain single engine.

Quality gates

  1. Exact baseline hash: full native P9.6 vs P9.7 PCM must be bit-identical with ROOM OFF or <=2m at 44.1/48k and 0/180°; front/rear 3.1m and 10m must show bounded low/center improvements. New Python verifier compares hashes, not nominal settings.
  2. Independently test bass, center-panned source, directional early-air, anti-phase L/R, sub/room mode finite bounds and state reset in 44.1/48k.
  3. Re-run existing P9/P9.1/P9.2/P9.4/P9.5/P9.6 full source tests, Windows native JUCE/ACK, SonKu Chromium and exact-SHA portable FAT.
  4. Owner headphone FAT required, including levels matched vs P9.6: 3.1m Signature, 7.5m Cinema, 10m Venue, source front/rear; assess kick impact, bass decay, singing articulation, frontal position, treble air, stereo separation, audible harshness/limiter activity. Reject P9.7 if it becomes louder but less external, or masks music.

Not approved for public release/merge or dearVR parity claims. See docs/SONKU_P97_STAGE_IMPACT_FAT_20261116.md.

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: fda0eef
Status: ✅  Deploy successful!
Preview URL: https://4d5a7bed.arsonmix.pages.dev
Branch Preview URL: https://feat-sonku-p98-signature-cin.arsonmix.pages.dev

View logs

Copy link
Copy Markdown
Owner Author

P9.7 selected-candidate source evidence at HEAD fda0eef669e0a59575c489e149662c1bf878c998

Fast full-source workflow 43822964118: Fast native contracts SUCCESS; SonKu P9 exact-source C++ PCM preflight SUCCESS.
SONKU_P97_CHECKPOINT_BIT_EXACT_AND_REAL_PCM_PASSED and SONKU_P96_INTEGRATED_SPECTRAL_AND_EXACT_LEGACY_PCM_PASSED.

  • Verified bit-for-bit identical complete PCM hashes against frozen P9.6 checkpoint for ROOM OFF, ROOM ON <=2m, ROOM ON at 10m with MIX=0%, each at 44.1/48kHz and front/rear. These are the critical rollback/no-regression guarantees; source / direct / early / modal / finishing stages are skipped at 0% MIX.
  • At 10m ROOM ON, level-matched-input full-graph 85Hz bass sinus-component amplitude is 1.113–1.124× P9.6, 1.9kHz center-component 1.127–1.141×, 7.5kHz air-component 1.003–1.016×. This is NOT a claim of 11–14% greater subjective bass or vocal volume.
  • At Signature 3.1m, full-graph bass 1.032–1.035×, 1.9kHz 1.034–1.037×, treble 1.000–1.004×. Stage is intentionally gentle there.
  • Isolated bands at both rates: bass impact 85Hz 1.260×, center input 1.9kHz 1.190×, early-reflection air 7.5kHz 1.132–1.135×. This confirms spectral program is not simply a makeup gain.
  • Kept P9.5 slow makeup and existing linked 0.98 peak guard intact; a bounded post-makeup, pre-limiter spectral finishing step prevents adaptive normalization from dimming cymbals when bass and center increase. Documented iterations rejected treble loss of ~13–16%; strict current candidate no-loss test now passes.

Remaining gates: exact-head Windows real JUCE GUI/native CTest/portable FAT, Chromium native editor FAT, and mandatory owner headphone/ASIO listening. Draft only. Not externally proven real vocal stem separation, individualized HRTF, or dearVR BRIR parity.

Copy link
Copy Markdown
Owner Author

Milestone A — ARCHIVED / CLOSED UNMERGED (no code merge or branch deletion). This P9.7 historical donor is preserved as the Venue golden reference, including immutable checkpoint branch and source SHA; the review PR is not a new merge instruction.

Immutable original head: fda0eef669e0a59575c489e149662c1bf878c998. Original source branch: feat/sonku-p95-p96-p97-stage-20261116; original PR target: feat/sonku-p95-p96-distance-reinforcement-20261116. Full 47-PR source-head inventory: docs/governance/MILESTONE_A_PR_BRANCH_RECONCILIATION_20261117.md on chore/milestone-a-pr-history-20261117 (planned for main). Closing this review surface is not proof of importing every experimental commit and is not permission to retune released SonKu or ImmerSIB. All branch refs, checkpoints and original code are preserved. Reopening requires a new, precisely scoped owner-approved reason.

@masarray masarray closed this Nov 17, 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