Skip to content

ArSon Vocal Throw M2 — real Music BUS auto-BPM + beat-phase synchronization #473

Open

Description

@masarray

Milestone B verified triage — 2026-11-17: P2 / vocal-throw / feature / backlog. Classification ≠ completion of product feature or physical FAT.

Scope / release boundary: Optional BPM Music BUS analysis; no fake beat phase.

Exit criteria:

  • Estimate BPM and phase from actual native Music BUS on worker thread with confidence/hysteresis and tempo holdover; no FFT/IO in ASIO callback.
  • Verify variable-tempo and stop/restart behavior and bounded precise readback to Vocal Throw inserts at 44.1/48 kHz.

Canonical issue ledger. Preserve historical evidence below; no speculative merging or rewriting approved audio.


Product goal

Automatic tempo detection from the actual music/backing BUS, not from the lead vocal microphone. ArSon Vocal Throw currently has real native phrase-end Manual/Auto/Hybrid and BPM/TAP, but no auto music tempo. This issue owns that remaining explicitly requested function.

Architecture / realtime contract

  • Add an explicit user-selectable tempo source (a live music/backing bus ChannelId), never silently sample MAIN or echo return.
  • Source PCM remains in the already authoritative native graph; a bounded nonblocking block-level/onset summary crosses to a dedicated analyzer off the ASIO audio callback. Do not run FFT/autocorrelation/model inference in the callback.
  • Analyzer estimates BPM and musical beat phase with confidence, half/double-time ambiguity protection, stable hysteresis, and holdover during intro/gaps. No tempo oscillation when backing stops.
  • A single versioned tempo transport/snapshot publishes accepted BPM/phase to participating VocalThrow inserts with device epoch and graph revision fencing. No second routing authority or secondary audio callback. VST3 latency compensation and MONITOR dry bypass stay intact.
  • Priority when source is unavailable/unreliable: explicit/song BPM or accepted TAP; never invent a tempo. UI must visibly distinguish Auto BPM locked, listening, and manual fallback.
  • Correct handling of tempo changes between throws without truncating active echo tails. Active tail retains its frozen BPM schedule; new tail uses next validated transport state.
  • Controls follow the existing approved /preview/ compact FX editor and accepted DSP ACK/readback; no browser DSP and no separate mixer page.

Acceptance

  1. Deterministic backing tracks at 72, 95, 120, 160 BPM (4/4, syncopated, sparse drums), 44.1/48/96 kHz, 64/128 sample ASIO callback: select backing BUS -> lock BPM (confidence/phase evidence) -> beat-aligned throws.
  2. Half/double-time, silent intros, held notes, tempo ramp, unpitched percussion and live vocal bleed do not cause unstable tempo jumps; clean fallback to TAP/manual.
  3. Source changes, WIRE retarget, SOURCE ONLY, bus mute/solo, device restart, sample-rate change and graph epoch rejection cannot leave stale auto BPM.
  4. No allocations, waiting, FFT, locks or logging in ASIO callback; bounded memory and SPSC overrun counters; CPU/xrun measurement vs Vocal Throw disabled.
  5. Manual/Auto/Hybrid phrase detection works independently of auto BPM; changing tempo never steals already sounding dual-tail phrases.
  6. Windows native CTest + Chromium /preview/ contract + physical singer/ASIO qualification before release.

Dependencies

M1 native Vocal Throw implementation in PR #469. Do not merge this issue into M1 test qualification or claim auto tempo exists until bus routing and analyzer are truly integrated and tested.

Activity

  1. masarray commented on Nov 8, 2026

    @masarray
    OwnerAuthor

    Progress 2026-11-09 — Music BUS Auto-BPM + beat-phase native PCM milestones

    Keep the issue OPEN and both PRs DRAFT. Remaining acceptance requires physical ASIO 64/128-frame 44.1/48/96k xrun/CPU, real musical syncopation / half-double-time stability, singer bleed, PDC, source/restart and practical listening. The 10ms onset window cannot be honestly described as sample-perfect beat detection. Do not merge PRs out of dependency order or confuse green CTest with live stage certification.

  2. masarray commented on Nov 8, 2026

    @masarray
    OwnerAuthor

    2026-11-09 — M3 stabilization and VST GUI checkpoint: draft PR #476 exact head 8dbf6d575cd8e37de5a5249743bdc86902bc1960 passed Windows MSVC/JUCE native 64/64 CTest (Actions #42529534785), including new strong-offbeat ambiguity/phase-lock recovery/no-false-steady-pad fixtures. Native editor Chromium desktop/mobile PASS (Actions #42529379402), premium VST UI and shared exact Close/Power chrome for EQ, Compressor and Vocal Throw. Full PR evidence. PR restored to its stacked M2 base and remains Draft. Keep this issue open until actual music + physical ASIO 64/128 frame xrun/CPU, live singer bleed and PDC latency evidence is available; a 10ms onset estimate is not sample-perfect detection.

  3. masarray commented on Nov 9, 2026

    @masarray
    OwnerAuthor

    ARSON VOCAL THROW — AUTHORITATIVE M1–M4 RECONCILIATION / SAFE HANDOFF

    Date: 2026-11-09 (Asia/Jakarta)
    Repo: masarray/arsonmix
    This is a progress-preservation checkpoint, NOT a merge, release or CI claim. All SHAs, branch ancestry and GitHub status were rechecked LIVE before posting.

    1) Authoritative milestone chain — DO NOT LOSE

    Stage GitHub PR / branch Exact preserved head SHA Live state Qualification
    M1 Native vocal FX #469 / feature/arson-vocal-throw-native-20261108 4db384851e6f6378cca4fd037b744c98bd1f0ba3 Open Draft, based on main, DIRTY (conflicting with newer main) Windows JUCE/MSVC + 63/63 CTest PASS run 42494827886
    M2 MUSIC BUS PCM Auto-BPM #474 / feature/arson-vocal-throw-music-tempo-m2-20261108 39ba9a48de8fe87270f9ed0cf9cac49fb004dfed Open Draft, based on main, DIRTY; contains M1 ancestry Windows JUCE/MSVC + 64/64 CTest PASS run 42499813920
    M3 Native sample-phase grid & detection stability #476 / feature/arson-vocal-throw-phase-grid-m3-20261109 8dbf6d575cd8e37de5a5249743bdc86902bc1960 Open Draft, CLEAN relative to M2 branch, not main Windows JUCE/MSVC 64/64 CTest PASS run 42529534785, Chromium desktop/mobile PASS run 42529379402
    M4 Hardware SVG/VST GUI #479 / feature/arson-vocal-throw-hardware-face-m4-20261109 5f6b1b01ba41a5cc559f0335d38b052967949f55 Open Draft, CLEAN relative to M3 branch, not main Fast UI contracts + real Chromium 1366×768 and 390×844 FAT PASS run 42548611390; screenshots artifact 13644975316

    Verified strict ancestry: M1 → M2 (+24 commits), M2 → M3 (+32), M3 → M4 (+30); each compare is ahead with behind=0. The current M4 head includes all three previous milestone histories, not just artwork. DO NOT cherry-pick M4 alone onto an incompatible main without M1–M3 native contracts.

    2) Latest main and reconciliation BLOCKER

    3) Feature checkpoint: native + UX

    • M1: realtime phrase-aware vocalist delay, Manual/AUTO/HYBRID, TAP, native accepted FX insert, frozen dual tails; native graph ACK and no audio callback heap.
    • M2: explicit MUSIC/BUS ChannelId PCM input tempo detector (not singer mic / not browser), bounded SPSC 10 ms energy/onset summaries, CONTROL BPM estimate (72/95/120/160) with confidence and manual fallback.
    • M3: one authoritative native PCM sample clock, epoch/revision fenced tempo transport, beat-phase confidence gate, octave/half-double-time hysteresis and coherent onset lattice, freezes tails on selected musical interval, no re-timing old tails; Manual/TAP when source stale. No duplicated callback.
    • M4 (latest accepted GUI direction): user SVG/Illustrator-derived black circular THROW hero without four directional markers, no big text inside; all seven knobs including MIX have the SAME body, stationary knob housing, pointer and outer 260° moving rounded cyan arc. Large LCD with inline-click BPM editing, positioned directly beside TAP. Symmetric compact desktop grid, stacked portrait flow, no footer clipping, minimal decorative text; preserve TAP aesthetic. EQ/Compressor/Vocal Throw share the SAME native-approved Close (24×24) and Power (26×24) CSS chrome and accepted DSP readback. All UI is on approved /preview/, not a parallel fake mixer.
    • Current M4 code touched only site/preview-v2/assets/mixer-native-vocal-throw.js, mixer-fx-rack.css, five SVG assets and tests/approved_native_vocal_throw_smoke.py compared with M3. M4 did NOT change native audio engine.
    • Original Illustrator 4974954.ai was inspected but is not committed. Lightweight derived SVGs, including a cropped source-lit icon embedded inside arson-vocal-throw-performance.svg, ARE committed; the UI can be rebuilt from the repository.
    • UI screenshot/geometry evidence: run 42548611390, artifact 13644975316. Tests measure THROW centerline, two encoder banks, LCD/TAP proximity, 7 knob arcs and stationary chassis, real rotary→native settings ACK/readback, source selection, screen BPM edit, Power/Close identity and mobile no-clipping. No untested browser-only DSP.

    4) EXACT safe continuation sequence

    1. Do not delete, reset, squash, or force-push any of the four milestone branches; record above SHAs as immutable recovery pins. Leave all PRs Draft until integrator proves release readiness.
    2. Work from latest main in a separate integration branch (or a carefully reviewed M1 rebase), preserve newest merged compressor/harmony/player/device/FX code. Resolve M1 conflicts semantically, not by blindly accepting ours/theirs; verify native bridge → dispatcher → PCM graph → callback → ACK.
    3. Integrate M2 atop resolved M1; integrate M3 atop that; integrate M4 last. Repoint stacked PR bases after each ancestor integration if appropriate. Never skip dependency stages, and don't consider mergeable=true vs parent proof against main.
    4. Run selective fast bridge/FX contracts and native Windows CTest on the NEW exact integration SHA once settled. Run targeted Chromium desktop/mobile GUI FAT after M4 integration; keep CI low-waste (avoid duplicate full Windows queue). If failure, fix before merge.
    5. Release qualification remains OPEN: physical ASIO 44.1/48/96 kHz at 64/128 samples, xrun/CPU, real singer mic and MUSIC BUS rhythm/syncopation/bleed, actual beat onset (10ms summary is NOT sample-perfect detection), FX latency/PDC, device restart, SOURCE ONLY/WIRE, song/session persistence, end-to-end native sound and usability. A green simulation CTest is not field proof.
    6. Protect the overall product UX promise: mixer graph / FX controls usable without requiring a hardware input first, no fake/unclickable GUI, no hidden monitor/arm gate, no browser illusion of DSP acceptance.

    5) State of this reconciliation

    No code, branch HEAD, PR base, installer or release has been changed; NO extra CI was dispatched. This issue comment is the durable cross-thread handoff; M4 PR #479 remains the latest reviewed GUI proof. Future agents should audit live main and all four PR heads again before touching conflicts, then continue from this exact chain. Do not mistake a checkpoint for a release or a merge.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions