Repository navigation
Conversation
…e and rhythmic echo DSP
… schema-versioned state
…g Windows CI target
… ChannelId engine
…nsitivity and native THROW ACK
… bypass and RT heap safety
…ompact desktop/mobile UI
|
/arsonmix-ci native a5e2565 |
2026-11-08 — Independent implementation / qualification auditVerified current branch head
Next qualification gates before marking ready: trigger the existing Windows MSVC/CTest path for this exact head; test physical ASIO 64/128 sample 44.1/48/96 kHz; validate real singer/breath/bleed/hold behavior, FX bypass and routing, Show restore and device restart. Product boundary: 'Auto' currently detects vocal phrase ending, not automatic BPM from music/backing bus. BPM entry/TAP exist; music-bus tempo/beat-phase engine and MIDI footswitch remain separate follow-up engineering work. No release claim until validated. Keep this PR as draft while these gates are outstanding; avoid claiming complete automated music sync. |
|
/arsonmix-ci native b2657cd |
|
/arsonmix-ci native 4db3848 |
Vocal Throw — native qualification recovery (2026-11-08)Exact previous Windows native run 42492553140 successfully compiled all JUCE/MSVC targets; CTest had 62/63 PASS. The only failing case was Root cause: the test expected a second auto throw after only 208ms of silence, but the detector's 18ms envelope release must first pass its -35dB gate before the 160ms silence counter. That interval was insufficient; this was a premature test expectation, not evidence that the callback crashed under normal audio. Test now verifies no early throw and a valid second throw after release + silence. Additional functional hardening: auto-then-late-manual now rejects a duplicate of the same ended phrase; the next sung phrase correctly resets the duplicate guard. C++ tests cover both trigger orders. Latest head: Remaining required hardware gate: physical ASIO 64/128 frames at 44.1/48/96k, xrun/CPU comparison and singer/breath/bleed audition. Auto BPM from isolated backing bus and MIDI footswitch are still not implemented; do not conflate Auto phrase-end with Auto music tempo. |
M1 qualified CI milestone / M2 handoff — 2026-11-08
|
|
Repository housekeeping (2026-11-10): Close as superseded, not merged. The M1–M11 Vocal Throw implementation was reconciled through merged #497 and the integrated M14 source through merged #499. This original pre-integration PR is no longer an independent merge target. Maintain feature/field qualification in #473 and #412; closing this historical PR is NOT physical ASIO certification. Original branch/PR history remains available for audit. Current main and any pending candidate must pass their own exact-SHA CI and hardware gates. |
ArSon Vocal Throw — native ChannelId live vocal FX
Implement actual C++ DSP in the authoritative JUCE PCM graph (not browser DSP or an independent audio engine).
Implemented
Boundaries and outstanding quality gate
Auto is vocal phrase-end detection; automatic tempo estimation from the separate music/backing BUS is NOT implemented yet. Current tempo is set by BPM or TAP. No MIDI footswitch binding yet. Energy-envelope heuristic is not trained singing/lyric phrase recognition.
This is an engineering qualification PR, not yet approved for main/release. Await exact-SHA C++ Windows MSVC/CTest, native WebView2/Chromium UI FAT, physical ASIO stress at 64/128 samples, real vocalist listening test (breaths/hold/bleed), routing and Show restore tests. Do not treat static/quick checks as live-singing certification.
Built from merged Compressor main SHA 8accbc4. Harmony PR #467 is separate and not altered.
Test strategy
Scope CI to existing quick Windows contracts and UI quick lane for PR; full native build/device qualification only for exact head, avoiding redundant workflow dispatches.
2026-11-08 — realtime/musicality hardening
Source hardening through
b2657cd0e6c56a69599162c56e6d81fd60ba23d5:Qualification: fast native contracts and targeted Chromium UI passed on exact
b2657cd; owner-authorized Windows MSVC/CTest build dispatched as Actions run 42492553140. Physical ASIO latency/xrun and singer/bleed listening trials remain unverified. Draft/release gate remains in force.Next architecture milestone (not falsely claimed implemented): tempo tracking from separate backing/music bus -> confidence and beat-phase/hysteresis -> lock-free shared transport -> sample-timed throw grid; optional phrase confidence and MIDI footswitch. Keep all heavy analysis off the realtime audio callback.