Skip to content

perf(harmony): P0–P2 realtime pitch, phase-safe analysis and rate-stable tone - #487

Merged
30 commits merged into
mainfrom
feature/harmony-p0-bounded-realtime-20261109
Nov 10, 2026
Merged

30 commits merged into
mainfrom
feature/harmony-p0-bounded-realtime-20261109

Conversation

@masarray

@masarray masarray commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

P0 milestone after Harmony #467 and optimization #485

  • Reconciled against current main 043748a552e66fd72e217cbf66aa35e6cc3e0b21 (Compressor Compressor P0: realtime scalar optimization + PCM parity and multi-instance diagnostics #484 retained, no shared-file conflict).
  • Includes perf(harmony): bounded pitch CPU optimization and synthetic callback profile #485's rolling energy and 48/96k synthetic CPU profiler, plus a time-sliced snapshot correlation engine: at most 32 lag candidates per DSP processing sample, rather than 175+ candidate burst every analysis hop. No RT allocation/locks, same one native ChannelId audio authority.
  • Preserve all existing GUI, SPSC/ACK, state schema, routing, Music/Manual guide, 2–5 voices and declared 1536-sample latency at 48 kHz. Fix voiced and shifter ratio smoothing to remain constant-duration across 32–192k device rates.
  • Add block-size parity regression (64/127/256) with nontrivial stereo vocal-like PCM, boundedness instrumentation and preserved dry PDC checks.
  • Document explicitly that changing 32ms lookback or calling granular shifting formant-preserving without real singer evidence is unsafe. The staged formant-aware shifter/quality–latency tradeoff and physical ASIO round-trip, xrun, 30min CPU and singer ABX remain release blockers.

Acceptance

Exact HEAD: fast native+UI, complete Windows 67+ CTest, one optional experimental FAT package; no duplicate CI runs. Keep Draft and do not claim production-ready until field ASIO and recorded-vocal quality gate.

Supersedes #485 after qualification; preserve its tested CPU changes.

P1 phase-resilient stereo analysis (2026-11-09)

  • Commit db33b3c66e6a0181a044356e1ccb06ef36f34318: added fixed-state StereoPhaseSafeFold to both vocal engine and read-only Music Track guide. Strongly anti-correlated stereo no longer erases pitch/chroma input; phase-coherent stereo retains original fold, 20ms evidence and 15ms smooth recovery, no new graph/router/PCM authority.
  • True native ChannelId PCM regression now exercises anti-phase A minor; 48/96k singer A3 and guide C major, zero-content Manual fallback and prior harmony PDC/block-parity tests.
  • Exact SHA fast PR CI 42569976663 success; complete Windows MSVC/CTest, FAT ZIP provenance verification and Inno Setup installer 42570084359 success.
  • Artifacts: ArSonMix-Windows-Installer-Preview-42570084359 and ArSonMix-Windows-FAT-Preview-42570084359. Both ZIP CRC verified after download; their embedded sourceCommit equals the exact head SHA.
  • Still Draft: no real singer ASIO physical FAT, no formant preservation or measured actual input-to-ear latency; declared 48k Harmony DSP lookback unchanged at 1536 samples.

2026-11-09 P0/P1/P2 deterministic DSP follow-up — b6e796891e6602d4d7d9b15a9e1c5bf8427c2914

  • P0 graph/PCM integrity: Proved antiphase analysis recovery cannot change the actual dry stereo waveform or the declared latency/PDC at both 48/192k (sample-exact regression, multiple ASIO-shaped block sizes). Still 1536 samples at 48k, not falsely reduced.
  • P1 timbre and sample rate: Added a RateStableColor lookup precomputed off RT. 48k settings keep legacy one-pole sound; 32/44.1/96/192k filters hold equivalent physical-time attenuation for the same Color value. Test checks six-sample/125 μs decay at 48k (not a too-long 10ms window that masks all differences).
  • P2 CPU and maintainability: Replaced five pow() computations per pitch analysis with 49 precomputed semitone intervals, one fractional base pow(), and bounded multiplies; saved a separate Color coefficient calculation per active voice. 42,600 semitone/detune ratio cases compared to the former formula within 1e-6. Pure C++ header, no new network/render owner or realtime allocations.
  • CI: PR fast native contracts 42577163398 passed on exact final SHA. One targeted Windows full CTest/FAT installer 42577277844 requested; results must be checked before approval.
  • Unresolved P0 release gates: Live singer male/female vocal timbre and formant preservation remain unproven; measure actual ASIO round trip and driver XRUNs with 2–5 voices before production approval. Do not conflate synthetic CPU CTest with hardware certification. GUI remains unchanged.

2026-11-09 D0/D1 — Natural Backing Vocal Depth (exact-source working candidate)

  • Product requirements: Natural Backing Vocal Depth
  • Architecture: Backing-Only Spatial Depth Engine
  • Research basis: direct/reverberant ratio, level, spectral depth, precedence effect (Nature 1999 / JASA / peer-reviewed spatial-hearing review). Do not equate stereo widening or a single echo with distance.
  • Implemented in 8c75946d0081a81f52d8a4b75b37e30078f95686: remove Wet-dependent lead attenuation (~3dB at Wet 45%); isolate backing-only spectral shading, lower backing direct-to-reflected ratio via existing Space, preserve per-singer early taps, introduce one fixed-size shared 4-branch Hadamard FDN with prepared 0.55s nominal decay. No user-facing extra mode, schema change or new audio callback owner; plugin PDC unchanged.
  • New exact-source CTest cases: lead formula independent of Wet, late-field impulse/decay and feedback bounds at 32/44.1/48/96/192k, mono polarity, restart clear, backing shading. Do not claim sound natural until acoustic listening and physical ASIO tests.
  • CI milestone: one targeted Windows MSVC/CTest/FAT run 42613364572 requested; check final result before marking installer qualified.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat ed11632

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: f02f35e
Status: ✅  Deploy successful!
Preview URL: https://05c3bc32.arsonmix.pages.dev
Branch Preview URL: https://feature-harmony-p0-bounded-r.arsonmix.pages.dev

View logs

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

P0 exact-SHA qualification — 2026-11-09

  • Candidate ed11632e5731f08d69d3b0150248165456cda20f based on current Compressor main 043748a5, incorporating prior perf(harmony): bounded pitch CPU optimization and synthetic callback profile #485 rolling-lag energy and synthetic profile.
  • Fast native contracts PASS (PR run 42566461328).
  • Full MSVC build and CTest PASS; FAT ZIP, provenance verification, installer build and artifact uploads PASS in exact-SHA Windows run 42566552615.
  • Artifacts: ArSonMix-Windows-Installer-Preview-42566552615 and ArSonMix-Windows-FAT-Preview-42566552615 (14-day retention per workflow). Users may retrieve them under GitHub Actions → run → Artifacts.
  • No GUI, routing, state schema or PDC latency changes. Sliced pitch correlation ≤32 candidate lags per DSP step; sample-rate-stable smoothing; callback partition determinism tests and previous native regressions remain green.
  • NOT release qualification: 48-kHz Harmony still reports 1,536-sample DSP latency; no tested formant-preserving shifter or physical singer ASIO listening/loopback/30min overload evidence. Keep Draft pending field FAT and acoustic quality comparison.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat db33b3c

@masarray masarray changed the title feat(harmony): P0 bounded realtime DSP and latency-quality qualification feat(harmony): P0 bounded realtime engine + P1 phase-safe vocal/music analysis Nov 9, 2026

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Verified P1 checkpoint (exact SHA db33b3c66e6a0181a044356e1ccb06ef36f34318)

Windows run 42570084359 all jobs SUCCESS: fast native contracts, MSVC full native build+CTest, GUI provenance, FAT ZIP packaging and validation, and per-user Inno Setup installer. Two artifacts uploaded and downloaded: ArSonMix-Windows-Installer-Preview-42570084359 (artifact 13653212875), ArSonMix-Windows-FAT-Preview-42570084359 (artifact 13653730811). Local downloaded ZIP CRC: PASS/PASS; both embedded provenance sourceCommit=db33b3c66e6a0181a044356e1ccb06ef36f34318.

P1 fixes a concrete L=-R cancellation defect in native voice pitch/grain source and other-ChannelId Music Guide while preserving dry stereo. Keep Draft until human singer/physical ASIO, latency and formant gates. Do not merge release claims from software-only tests.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat b6e7968

@masarray masarray changed the title feat(harmony): P0 bounded realtime engine + P1 phase-safe vocal/music analysis perf(harmony): P0–P2 realtime pitch, phase-safe analysis and rate-stable tone Nov 9, 2026

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 P0/P1/P2 exact-SHA qualification checkpoint

  • Commit: b6e796891e6602d4d7d9b15a9e1c5bf8427c2914, parent from the previously successful P0/P1 Harmony candidate; current main independently verified as 043748a552e66fd72e217cbf66aa35e6cc3e0b21.
  • Code: rate-stable Color coefficients via control-thread LUT, precomputed +/-24-semitone ratios (5 less pow per detected pitch hop), one color coefficient lookup per active PCM sample; 48k legacy parity, regression for dry anti-phase PCM at 48/192k, seven Color values across 32/44.1/48/96/192k and 42,600 ratio comparisons.
  • CI: exact-SHA fast PR run 42577163398 SUCCESS; full Windows run 42577277844 SUCCESS (MSVC native build/CTest, approved GUI provenance, FAT package validation, Inno Setup installer).
  • Artifacts: ArSonMix-Windows-Installer-Preview-42577277844 (artifact 13656677217, ZIP size 5,555,244 bytes) and ArSonMix-Windows-FAT-Preview-42577277844 (artifact 13656870710, 5,699,492 bytes). Both downloaded ZIP archives have CRC PASS and exact sourceCommit=b6e796891e6602d4d7d9b15a9e1c5bf8427c2914 in respective provenance JSON.
  • Release qualification: Remains Draft because this work does not deliver low-latency formant-preserving resynthesis or observed physical Windows ASIO driver/singer evidence. Preserve 1,536-sample Harmony DSP PDC at 48 kHz until acoustic data proves a safe reduction.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat b83597b

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 P0 realtime slice + hostile PCM milestone — exact HEAD verified

Commit: b83597bd26e7583c8dc135a9cc6ac2f8fdbcd43e; main remains b630cf8be8c38be64853a5070a2e254684eb2de8, PR is intentionally Draft.

Implemented: Reduced periodic autocorrelation work from 32 to <=8 candidate lags in a single sample, with analysis cadence sped from 8 to 32 slices per detection hop to avoid dropping frozen PCM windows. Added audio-owner completed/abandoned search counters and 32/44.1/48/96/192 kHz × 64/127/256 block regressions. Independently enumerated all integer sample rates 32,000..192,000 Hz: at least 62 samples of search-completion headroom (mathematical schedule, not measured ASIO timing). Stereo analysis-only fold now bounds its actual returned sample; hostile finite FLT_MAX, NaN and +/-infinity are safely contained before the voice's histories. No GUI, graph authority, VST routing, state packet or 48k DSP PDC change.

Qualification: Fast native checks run 42581079521 PASS; Windows MSVC, native CTest, FAT integrity and Inno Setup run 42581191745 ALL SUCCESS. Downloaded artifact ZIP CRC PASS for both. Embedded installer and FAT provenance sourceCommit both equal b83597bd26e7583c8dc135a9cc6ac2f8fdbcd43e.

GitHub installer artifact: ArSonMix-Windows-Installer-Preview-42581191745 (ID 13658290036, 5,556,724 bytes ZIP; contains ArSonMix-Windows-x64-Setup-Preview.exe).
GitHub portable hardware FAT: ArSonMix-Windows-FAT-Preview-42581191745 (ID 13657632416, 5,699,587 bytes ZIP).
Not completed: formant-preserving pitch synchronous engine; safe shorter-than-1536-sample 48k PDC; real singer/ASIO latency and driver XRUN field data. Do not merge as production release until these P0 gates are addressed.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 770f390

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 770f390

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 field screenshot — Harmony GUI cannot control live DSP (P0 bridge recovery)

Reproduced root cause from screenshot and source: UI footer said DEVICE / GRAPH CHANGED · REOPEN ACCEPTED FX. mixer-native-harmony.js froze when the accepted ChannelId/slot/instanceId ceased to match an open editor's cached identity. User clicks were rejected from send(); no native mutating action was emitted; the readback loop never re-established a valid DSP identity. PCM continued because its native callback was independent.

Fix 770f3908e28dabb250dfb0a687e573b7869dfa51: acceptedLocation() locates current approved snapshot and exact insert identity on the same stable ChannelId, follows FX slot moves and safely recognizes a recreated Harmony at the formerly selected slot. reconcileBinding() automatically invalidates only stale in-flight ACK/readback, refreshes native identity and requests a NEW authoritative DSP snapshot on the existing console tick. A bounded latest-user-field patch is held during transient disconnection and replayed only AFTER accepted snapshot; never sends to a different channel or invents PCM. UI handles unavailable FX without requiring close/reopen. Fixed editor slot label on reorder and initial transitional admission. No DSP, PCM callback, routing or GUI skin changes.

Tests: New real Chromium approved_native_harmony_smoke.py scenarios simulate revision change, new same-channel Harmony instance, old out-of-order ACK, insert reorder, temporary FX removal/restoration, held user edit and latest audio ACK/readback. Updated structural assertion to require the accepted-graph binding rather than the old initial-only gate. Fast UI 42593462389 SUCCESS; full real Chromium across published UI + Harmony 42593573980 SUCCESS. Native quick 42593462603 SUCCESS. One exact-SHA Windows MSVC/CTest/FAT packaging 42594272796 running; do not label installer qualified until final result.

Separate audio-quality defect: robotic voice is not resolved by this bridge fix; granular harmony shifter lacks formant preservation and 48k PDC is still 1536 samples (32ms). It needs independent resynthesis/formant and ASIO physical acoustic verification; do not equate GUI test success with global top-quality singer FX.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 4fd5d63

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 P0 Harmony recovery — native resynthesis and inspectable vocal A/B milestone

Exact candidate HEAD: 94f768d2df9c56431d78d2044d5bc9229e2877ab (Draft; not merged).

Actual code delivered:

  • New engine/builtin_fx/HarmonyPitchSynchronous.h: bounded causal source-period-aligned overlap-add resynthesis candidate. Reads original vocal samples at unity speed and adjusts synthesized mark spacing rather than blindly resampling 16ms grains. Runs in existing ChannelId PCM callback, no heap/mutex/GUI/routing changes, 1536 samples of declared latency @48k unchanged. Correlation confidence gates a 12ms crossfade to retained granular fallback for uncertain/nonperiodic material.
  • New tools/harmony_offline_ab.cpp with deterministic native WAV renderer: same input -> original lead reference, legacy grain mix, candidate Space .6, backing-only difference, mono, Space 0 and Space 1 + metrics.json for independent singer A/B. Offline-only switch to original shifter; no user-facing DSP lab mode or serialized session change.
  • New tests/harmony_pitch_synchronous_test.cpp, CMake integration, and path-scoped .github/workflows/harmony-fast-dsp.yml to actually compile/run PSOLA, complete Harmony PCM engine and A/B renderer C++ on every relevant PR update.
  • Docs updated: docs/ARSON_HARMONY_NATURAL_ENSEMBLE_ARCHITECTURE_20261109.md.

Verified checks: exact-head Linux C++ native tests (all three) PASS 42627679146; fast approved UI and baseline native contracts PASS on same SHA. Windows full MSVC/CTest + verified FAT and Inno installer is running as one SHA-pinned canonical workflow 42628304651, triggered without changing this candidate via one-shot CI branch 42628287177. Do not declare Windows result before completion.

Still P0 unqualified: source positive peaks are approximate waveform epochs, not proven glottal closure detection; retained grains/soft fade are engineering fallback, not certified formant preservation or natural vocal quality. No genuine adult singer male/female vowel/fricative/slides A/B performed here, no F1/F2/F3 report, no 30-min 48/96k ASIO xrun/roundtrip/CPU proof. Installer if successful is an experimental FAT only. Do not merge or claim global-class human voice until acoustic and physical gates; do not overwrite concurrent Player/Compressor/Throw work on main.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 exact-HEAD Windows qualification COMPLETE — native CTest + Inno FAT

Source SHA 94f768d2df9c56431d78d2044d5bc9229e2877ab
Canonical Windows run: https://github.com/masarray/arsonmix/actions/runs/42628304651 — SUCCESS
MSVC x64 build: SUCCESS. CTest: 69/69 PASS, 0 failed. Includes NativeHarmonyPitchSynchronousFormantCandidateAndBlockParity and NativeHarmonyOfflineDryAndABRenderHarness. Source SHA verified before build.
Portable ZIP/GUI verification: SUCCESS, FAT_RELEASE_ACCEPTED. Installer first install, running-app graceful update, SHA-matched PE and uninstall/private-Studio checks: INSTALLER_RC_VERIFIED.
Published inspectable artifacts (preview, retention 14d):

Still DRAFT and acoustically unqualified. This is a clean native engineering / source and packaging milestone, not evidence of a global-class human vocal sound: real adult singer A/B, vowels F1/F2/F3, breath/fricative intelligibility, pitch-slide expression, and physical ASIO 48/96k XRUN/roundtrip/callback p95/p99 remain outstanding. No merge, main modification, production release or hardware certification implied.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 D2 — Virtual backing-stage spatial enhancement (FIELD listening follow-up)

Field input: previous pitch-synchronous Harmony is noticeably less robotic and already sounds reasonably natural. Remaining request is perceptibly greater spatial distance/3D separation between an intelligible forward lead and three musical backing singers, not more artificial pitch processing.

Exact candidate HEAD: 604b82223db2471ee772e9c5c013720515b9fc71 (PR remains Draft).

Source implementation

  • engine/builtin_fx/HarmonyStageSpatial.h: fixed 0.22ms contralateral precedence cue per singer with Space-dependent level crossfade, not delay-time modulation or phase inversion. No per-callback allocations, FFT, host/device calls or audio ownership change.
  • HarmonyEnsembleSpace.h: retains existing first 12–20ms early reflection, adds a second independent per-singer 27–38ms early reflection using the same per-singer ring expanded to 8192 samples (valid at 192kHz).
  • HarmonyDsp.h: backing-only direct/early/late gain scene: at Space s direct 1-0.62*s, early 0.49*s, late 1.08*s; the ONE shared pre-existing backing room remains, and the dry stereo lead, natural pitch-synchronous resynthesis, native ACK/graph/PDC and user Space knob are unchanged.
  • Offline-only exact previous stage routing retained for old/new A/B with same new pitch engine. tools/harmony_offline_ab.cpp now exports 08_previous_depth_space_060.wav and 09_previous_backing_only.wav, besides the new stage and front lead WAVs.
  • tests/harmony_stage_spatial_test.cpp checks 32/44.1/48/96/192k, three/five singers, first/second reflection taps, monotonic early/direct-energy trend vs Space, positive mono energy, 48k 64/257 block parity, 1536-sample declared PDC, and exact Wet0 lead invariance.

Qualifications

  • Exact-head focused native C++ CI 42633744794: PASS. Prior spatial source commit fast native 42633504110: PASS.
  • Current canonical Windows MSVC/CTest + Inno installer FAT #42633886160 running on pinned HEAD. Do not claim Windows success until final checks.
  • This is stereo spatial depth, not an individualized binaural HRTF or true coordinate-based 3D audio. Physical ASIO singer hearing, headphones + nearfield + mono, matched-level old/new depth perception, intelligibility and driver xrun/CPU/latency are still unverified. Keep Draft and do not change main until integrated with Player/Compressor and those qualifications.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

D2 Windows-specific CTest regression found and corrected (exact head 94a4039)

The first full Windows run #42633886160 compiled successfully but had 69/70 CTest PASS, with a SegFault in NativeHarmonyMusicFromOtherChannelIdAndStateMigration. This was NOT treated as a green build; no installer was published.

Root cause from source + comparison with the previously green fixture: tests/harmony_music_bank_test.cpp heap-owned the first live processor but still constructed TWO additional ChannelHarmonyProcessor restore/migration fixtures on the default Windows test stack. Each processor embeds the now larger fixed-capacity Harmony DSP (extended per-singer reflection rings), so two simultaneous ~large processor stack locals can exceed the Windows stack budget before assertions. The DSP is heap-owned by the production FX graph, not placed on its audio callback stack. The same project previously encountered oversized stack-local Harmony fixture failures in tests/harmony_dsp_test.cpp.

Repair: commit 94a4039fa18c1fecec91efdb8f356cc083b5386b heap-owns both restoration fixtures through std::make_unique<ChannelHarmonyProcessor>, keeping their entire original JUCE/ChannelId/PCM assertions active. No silencing, skipped test, graph reroute, audio callback allocation or pitch/spatial change.

After fix: exact-head native C++ fast #42636728287 PASS; fast UI/contract checks PASS. One source-pinned complete Windows MSVC/70 CTest + Inno FAT requalification is now running: #42636832559. Do not use the older failed build or claim a new installer until its full result is confirmed.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

2026-11-09 Harmony P1 — singer-reviewed musical stage-depth refinement (exact SHA c0cd439f545724e9423102fd3ff7063028a36600)

User field feedback: current pitch-synchronous vocals already sound noticeably more natural / not robotic. Prioritize subtle musical three-singer front/back separation; do not disturb resynthesis.

Incremental native changes on top of the already present BackingStageSpatial and two-reflection/shared-room architecture:

  • 2230cd9: rate-stable (4.1kHz) single-pole far-ear-only spectral shadow in the existing 0.22ms fixed-delay stage ring. Near-ear voice remains unfiltered; no new buffer, routing, native latency or knob. New 350Hz/8kHz far/near response tests across 48/96/192k, Space0 transparency.
  • c0cd439: phantom-center backing singer modestly recessed (down to 0.83x direct at Space1) relative to its unchanged shared room send and early reflections; only wet backing affected, not dry lead. Tests assert exact Space0 parity and strictly retained center room feed. Side voices remain moderately lateral; no anti-phase widening or fake HRTF location.
  • Existing tests/harmony_stage_spatial_test.cpp covers DRR direction, all supported rates, stereo, mono, exact callback 64 vs 257 partition, lead Wet0 and PDC; native fast QA PASS 42638162437. UI smoke PASS 42638162463. Canonical pinned Windows FAT installer qualification is running 42638327447; do not claim Windows PASS until complete.

No production merge or GUI alteration. Still needs headphone+monitor and mono subjective A/B with real singer to judge whether stage distance improved without sibilance smearing, plus physical Windows ASIO xrun and roundtrip evidence.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Verified D2 spatial-stage Windows baseline (exact SHA 94a4039)

Canonical Windows run #42636832559: SUCCESS. The SHA-guarded source 94a4039fa18c1fecec91efdb8f356cc083b5386b built MSVC x64; 70/70 CTest PASS including the previously crashing native Harmony Music Channel migration test. Portable source/GUI/PE: FAT_RELEASE_ACCEPTED; Inno first install, running-process update, uninstall & Studio preservation: INSTALLER_RC_VERIFIED.

Qualified experimental artifacts at this exact baseline SHA (14-day GitHub retention):

Important live-head distinction: Harmony branch advanced afterward to c0cd439f545724e9423102fd3ff7063028a36600 with additional contralateral spectral shadow and center-back direct recession, both in HarmonyStageSpatial.h. Focused native tests on that newer SHA #42638162437 passed; the old verified installer above DOES NOT contain those newer two commits. Another source-pinned full Windows run #42638327447 is already underway for new HEAD. Do NOT trigger a second duplicate run. Do not claim 70/70 for new HEAD until that exact run finishes. No production/field ASIO spatial listening certification.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

D3 — Musical Room Focus, exact source d9e7b79e223962d15905def469247aa42ef26593

User acceptance: existing Harmony pitch sounds much less robotic. Preserve pitch-synchronous lead and focus on musical back-row depth.

Implemented in DSP, not GUI paint:

  • New pure C++ engine/builtin_fx/HarmonyRoomFocus.h on the ONE shared backing room send (170Hz first-order HP, prepared per sample rate). Lead PCM, backing direct, initial early reflections, VST3 host, source graph, output and fixed latency remain unchanged.
  • Gentle phrase-dependent late-room blend driven from the latency-aligned dry lead: 12ms attack / 240ms release; room-return reduction capped at 22% only at Space=1 and long strong lead, then blooms naturally in short lead gaps. No compressor on singer, no extra knob or saved state, no heap/lock/realtime FFT.
  • Single existing Harmony stage and ONE shared FDN retained. Offline WAV A/B now exports 10_room_focus_off_space_060.wav, 11_room_focus_off_backing_only.wav using identical source/pitch/stage; dry Wet0 is bit-exact against both reference choices.
  • New native sample-rate scoped tests at 32/44.1/48/96/192k for 75Hz vs 600Hz cleanup, bounded program-dependent gain attack/release, Space0 unity, restart state reset, plus existing 64/257 block parity, mono, stereo, exact fixed PDC, real native DSP render.
  • Exact source GitHub Actions: Harmony fast C++ DSP PASS, native Windows PR checks PASS, UI smoke PASS. One pinned MSVC x64 CTest + FAT/Inno installer workflow running 42646357779; qualify only after confirmed success/artifacts.

Still Draft. No real human singer blind A/B, rated back-of-stage perception on headphones/nearfield/mono, or physical ASIO XRUN/p95/latency data. This is a conservative audible candidate, not an HRTF or certified live-performance claim. Do not merge into main until release gates and cross-thread engine reconciliation.

…sor main

Two-parent integration preserves original Harmony history and current main Player/Compressor. CMake merges tests and dependencies additively. Retain PR Draft pending singer listening and physical ASIO.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Harmony D3 reconciliation checkpoint — 649409a

Performed an explicit two-parent integration on PR #487 branch only (not on main; no forced rewrite).

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Harmony D4 user playback fix — sweet-spot WET presence + dial label spacing

Owner screenshot showed 4 voices, WET 100%, Space 60%, backing still too thin despite human-sounding timbre; knob captions COLOR/SPACE/WET/GAIN touch the arc. Exact branch HEAD: afa6f2f432ebadb15335db99a0f590b9662df99b (based on reconciled two-parent Player/Compressor main merge 649409a0).

  • Root-cause DSP: previous per-singer gain 0.65 * Wet / sqrt(voices) gets spatial direct attenuation afterward. New control-thread-only voice amplitude Wet * (0.77+0.38*Wet) / sqrt(voices) (approx +3.2dB at Wet45 and +4.95dB at Wet100); dry lead not ducked or boosted, original pitch-synchronous/Space staging, mono policy, 1536-sample @48k PDC unchanged; no realtime new math. 2–5 voices remain count-normalized. No hard limiter/real singer headroom claim.
  • Added original-vs-new WET-only offline native reference switch before processing, and 12_presence_old_space_060.wav, 13_presence_old_backing_only.wav matched A/B stems. Native C++ regressions assert no wet leakage at zero, monotonic/bounded gain through 5 singers, actual 4-voice/Wet100 backing stem RMS >1.5x previous and <1.9x while Wet0 lead remains sample-exact.
  • Visual geometry: previous flex gap 1px ignored the dial pseudo ring's inset:-11px; added only margin-top:12px;line-height:11px on four captions without larger knobs/panel. Native Harmony real Chromium verifies clearance, numeric output, identity, key/ACK, and responsive editor at 1366/720/390 viewports.
  • CI proven this exact SHA: Harmony fast native DSP PASS, UI fast PASS, Windows PR fast PASS, native Harmony Chromium screenshot QA PASS with screenshots artifact. Exact-SHA Windows FAT/Inno 42655553399 launched; do not mark qualified until completed.
  • Keep Draft: user must listen to actual singer 3/4/5-voice level balance, clipping/headroom, expression and physical ASIO driver xrun. Only merge once reconciled with newer main/active workstreams and validated.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Harmony D4 final UI+DSP reconciliation checkpoint — exact SHA f02f35eee61d601fdaa2bb59ba7c4ca5826330f2

Screenshot-driven fix refined after D4 first patch:

  • Musical WET presence restored in DSP (control-thread Wet*(0.77+0.38*Wet)/sqrt(voices), ~+3.2 dB at Wet45 and ~+4.95 dB at Wet100 vs previous), dry lead and pitch/Space latency unchanged. Native quick C++ DSP, offline native 4-voice backing-stem RMS A/B and 2–5 voice control coefficients PASS (42656198566).
  • Four COLOR/SPACE/WET/GAIN captions now clear inset:-11px glowing ring. Follow-up layout commit redistributes vertical whitespace rather than enlarging the editor: desktop row-gap 14px / padding 3px, narrow row-gap 10px / padding 2px. Added inter-row geometry assertions to prevent a caption fix moving numeric values into the NEXT knob's halo.
  • Exact-head real native Harmony Chromium test across 1366×768, 720×600 and 390×844 PASS, including actual caption-to-own-ring and numeric-to-next-row ring clearance, output value/caption separation, ACK, Music/Manual and single editor (42656291500); screenshots artifact.
  • Windows FAT/Inno/CTest exact-source run 42656432831 launched after Chrome passed. NOT yet qualified until it succeeds. Prior fully compiled reconciliation 42651518008 PASS at previous parent SHA.
  • Branch remains 30 commits ahead, 0 behind current main and Draft. Main Player M1 and Compressor restart changes retained; all other active PRs remain separate. Field singer loudness, wet max clipping margin and physical ASIO xrun still require owner evidence.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

CHECKPOINT / HANDOFF — Owner field defects after Harmony D4 listening (2026-11-10 WIB)

Owner audio acceptance (preserve): Vocal Harmony pitch/voice is now considerably more human, with D4 backing WET presence and stereo depth judged "lumayan bagus". Do not undo this acoustic baseline.

Two new P0 defects reported by owner in actual application:

  1. Repeated Harmony FX ON/OFF causes an audible impulse/jump/crackle between original voice and processed Harmony; must transition seamlessly, without muting MAIN, changing graph/device, resetting FX session, changing published PDC, or adding latency.
  2. Music Key mode sometimes produces crackles while operating; manual mode is baseline. Distinguish sample discontinuity from voice re-harmonization versus actual ASIO underrun, and prove source-channel music key switches/fallbacks have musical click-free transitions without stale source/identity.

Live GitHub refs verified for handoff: Harmony PR #487 remains DRAFT at exact approved D4 checkpoint f02f35eee61d601fdaa2bb59ba7c4ca5826330f2. Native Linux DSP, UI smoke, real Chromium 3-viewports and MSVC x64+all CTest+FAT/Inno preview run 42656432831 SUCCESS. Main since advanced one commit to 9bd81a1b8da764a2f4523e70c0db60f583aa291f (independent Player KEY/TEMPO). PR #499 1dc70dc11185f29b4821a4765f68579b85136f33 is DRAFT VERIFICATION ONLY — DO NOT MERGE, combining ASIO4ALL, Harmony D4, Player. Reconcile freshest branch and main, avoid silently merging other drafts.

Source-backed P0 focus (not yet root-cause proof):

  • engine/core/LiveChannelFxGraph.h::FxInstance constructor calls bypassFade.prepare(false, processor->getSampleRate(), 0) even though it separately captures latencySamples = processor->getLatencySamples() and prepares dryDelay. FxBypassEnvelope.h has distinct zero-latency wet/dry crossfade versus latency-bearing fade-through-silence path. Determine correct phase/PDC policy for 1536-sample @48k Harmony (possibly aligned dry and wet), and why abrupt switching is audible. Never blindly change 0 to latency without verifying PDC and acoustics.
  • FxInstance::render skips processor->processBlock after fully bypassed and copies delayed dry; when re-enabled, pitch-sync, detector and shared-room DSP may contain aged/frozen PCM/tail. Test first full block and 32ms reacquisition. Scalar bypass updates through DynamicChannelFxLeaseBank::setInsertBypass and setRackBypass should not reconstruct graph or reset state.
  • ChannelHarmonyProcessor.h::observeMusicTrack calls guide_.process, then dsp_.setGuideKey on source ChannelId callback; HarmonyDsp::setGuideKey resets currentNote_/candidateNote_/noteHold_/voiced_ when effective key/scale changes. HarmonyDsp::finishPitchAnalysis recomputes pitch ratios and can pitchSync_.reset for >25% source-period jump. Direct Music callbacks via DynamicChannelFxLeaseBank::operator() after selected channel FX. Verify periodic key transitions/fallback and vocal source voice re-entry for audible discontinuity; check whether 36-bin detector evaluation bursts harm realtime deadlines (measure, do not assume).
  • HarmonyMusicKey.h chroma evaluates at ~0.1s hops, minimum ~1.2s, candidate hold and 1.5s uncertain fallback; ensure hysteresis and continuity across lost/ambiguous source and song key changes without permanently holding wrong harmony.
  • Retain wet presence Wet*(0.77+.38*Wet)/sqrt(voices), lead-independent dry PCM, rate-correct stage/room, LatencySamples and existing GUI contract. No new recovery controls or user action.

Required evidence / acceptance (developer thread must IMPLEMENT, not just propose): Native exact-input OFF/ON/OFF repeat at arbitrary sample offsets and 64/128/256 chunks, three/four/five voices, 48/96k minimum, 192k stress; same pre- and post-FX dry PDC aligned, no spike, drop-to-zero, tail burst, stale echo; rapid toggles reversals and rack OFF preserve individual FX state. Music source stable key, transitions C major/A minor, ambiguity, silence, disappearance, retarget and return, while vocal keeps singing: no click or drop, no stale Chapter/ChannelId, no accidental audible music WIRE. Automated PCM derivative/transient and spectral click evidence with comparison to prior baseline; finite/non-clipping, deterministic block-size independence, fixed latency. Measure DSP p50/p95/p99 and real ASIO4ALL driver XRUN counters separately, excluding false acoustic vs xrun classifications. No malloc, IO, locks, full-spectrum scans or logs in RT callback; one native PCM graph, no new audio thread. CI selective, exact SHA Windows CTest and FAT installer only after stable targeted tests; field user listening and physical ASIO required before promoting production.

Action: Another thread should audit live refs, reproduce two mechanisms, fix architecture (not compensatory UI tricks), add deterministic regression fixtures and report PR/commit/tests/artifacts. Keep #487 draft and protect Player/Compressor/ASIO cross-thread work until evidence.

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