Paul-Michael Sorhaindo

← all projects

pmsorhaindo/ModellingAudioProcessors private repo · project board · as of 2026-11-20 evening (Europe/Amsterdam)

ModellingAudioProcessors

What it is

Research and C/C++ implementations of analog audio processor models. The first target is a fast valve saturator plugin (AU / VST3) inspired by the Thermionic Culture Vulture. It is not an emulation of that unit, and nothing in the repo is measured from one. No schematic is public, so the circuit values are example values (prototypes/fullstage/dk/circuit.py says where each one comes from). The product name will not use a brand or device name.

The repo is https://github.com/pmsorhaindo/ModellingAudioProcessors (private). The project’s own code is MIT. Third-party SDKs keep their own licences. Nothing is signed, notarised, published or released.

Rules, from the README:

Every pull request is reviewed by a different model from the one that implemented it (Claude / Codex / Antigravity / Cursor cloud agents). PRs open as drafts; macOS CI jobs run only on ready-for-review PRs and manual dispatch, not on drafts or pushes to main, because macOS minutes bill at 10× on this private repository. AGENTS.md: never merge without Michael’s approval.

Goals and direction

The current product direction was decided by Michael on 19 Nov 2026, on https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/75:

The first loadable plugin, merged in https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/91 (closing #22) on 18 Nov 2026 at 10:48 Europe/Amsterdam, wraps the validated stereo 12AX7 dkcore::StereoStage. It has no custom GUI; the host shows its generic parameter UI. Parameters: Drive, Bias, Output, Quality (Eco 4× / Standard 8× / HQ 16×) and Bypass. The mapping from host full scale to the stage’s volts is still provisional.

The 18 Nov 2026 README status (written the same day as several later merges) listed a 12AX7-stage model, aliasing checks, the C++ core, CPU figures, a Python 6AS6 DK stage, the first plugin build, and a WDF reference. Several items that README still called “not done” were closed later that day or the next: stereo vector maths (#67 / PR #68), the x86-64 stereo benchmark (#76 / PR #94), host sample rates other than 48 kHz (#74 / PR #95, completed), and parameter smoothing (#73 / PR #97, completed). Treat issue state as authoritative where it is newer than that README paragraph.

Workstreams on the board:

Architecture

Drawn from the README structure table and plugin/README.md. The plugin does not depend on the WDF tree. Circuit values are example values, not from a commercial unit.

flowchart LR
    classDef local stroke-dasharray: 5 5

    Refs["references/ manuals, datasheets, papers"]
    ValveLab["prototypes/valvelab/ static valve-stage lab"]
    Fit["prototypes/fit6as6/ 6AS6 pentode fit to Raytheon 1960 curves"]
    Research["prototypes/research/ grid current, panel controls, datasheets"]
    FullPy["prototypes/fullstage/ Python nodal DK stage + ngspice references"]
    Core["prototypes/fullstage/cpp/ header-only C++17 realtime core, 12AX7"]
    Plugin["plugin/ Gyre: AU v2 + VST 3 wrapper of StereoStage"]
    WDF["prototypes/wdf/ WDF reference only, not the shipping core"]
    SixPy["prototypes/fullstage/dk6as6/ Python 6AS6 DK stage — not yet ported to C++"]

    Refs --> Fit
    Refs --> Research
    ValveLab --> FullPy
    Fit --> SixPy
    Research --> SixPy
    FullPy -->|port of the 12AX7 stage| Core
    FullPy -.->|compared with| WDF
    Core --> Plugin
    SixPy -.->|"C++ port is #104, not done"| Core

Realtime processing in the current plugin, from plugin/README.md. The volts mapping is marked provisional. Bias range ends 0.5× and 2.0× have not been validated beyond building a pack and solving the DC point.

flowchart LR
    In["Host audio"] --> Drive["Drive 0–8×, 20 ms ramp, before the stage"]
    Drive --> Map["Host full scale to volts — provisional"]
    Map --> OS["Oversampling: Eco 4× / Standard 8× / HQ 16×"]
    OS --> Stage["dkcore::StereoStage, 12AX7 nodal DK"]
    Stage --> Out["Output 0–4×, 20 ms ramp"]
    Out --> Host["Host output"]
    In --> Bypass["Bypass, 10 ms crossfade to dry delayed by latency"]
    Bypass --> Host
    Bias["Bias: 0.5–2.0 × Rk, 20 ms ramp in the core"] --> Stage

6AS6 path toward a verified Gyre v1 model

Dependency order recorded by Michael on #75 (19 Nov 2026, 20:20 Europe/Amsterdam), with later issue state. How T / P1 / P2 are actually wired is not stated in any public source; #111 says nothing should be invented and passed off as measured. P2 cannot be supported until https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/123 (refit the g3 cut-off region of the device fit).

flowchart TB
    i102["#102 Circuit values — completed, PR #118"]
    i103["#103 Post-distortion filter — still open; PR #119 merged a TPT SVF prototype"]
    i110["#110 Bias as cathode voltage, 0.1–0.6 mA — completed, PR #120"]
    i111["#111 T / P1 / P2 wiring — in progress, PR #121"]
    i112["#112 Decide on unvalidated g1 grid current — Todo"]
    i113["#113 Aliasing and eco / quality for the 6AS6 path — Todo"]
    i123["#123 P2: refit g3 cut-off — Todo, P1; blocks P2 support"]
    i104["#104 C++ port of the 6AS6 stage — Todo"]
    i114["#114 Design the 12AX7 swap-in in Gyre — Todo"]
    i115["#115 Parameter smoothing and mode switching — Todo"]

    i102 --> i110
    i102 --> i111
    i102 --> i112
    i102 --> i113
    i102 -.-> i103
    i110 --> i111
    i110 --> i113
    i111 --> i113
    i111 --> i123
    i123 -.-> i111
    i110 --> i104
    i111 --> i104
    i113 --> i104
    i104 --> i114
    i104 --> i115

Uncertain or blocked:

Related, not on that critical path: https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/90 (lane-math, still a 12AX7-path question until #104 exists) and #109 (reusable tooling). Follow-ups opened from PR #121’s review: #123, #124 (burst recovery-time metric), #125 (TIGHT ngspice records), #126 (ngspice hang under Rosetta; move to native arm64 ngspice), #127 (macOS fullstage CI headroom).

Board snapshot

User project https://github.com/users/pmsorhaindo/projects/7, all 36 items, as of 21:07 Europe/Amsterdam on 20 Nov 2026. (An earlier first page of 30 was completed by a later --limit 100 fetch after a temporary Projects GraphQL rate limit.) Priorities are filled on #123–#127 only (P1: #123, #124, #126; P2: #125, #127); earlier items have no Priority field in the snapshot.

Board: github.com/users/pmsorhaindo/projects/7 · 36 items

# Type Title Status Priority Labels
Loading board…

Issue dependencies

See the 6AS6 path diagram above for the v1 model sequence. Other documented links:

flowchart TB
    ws10["#10 C++ core and plugin"]
    ws11["#11 Octatrack port"]
    i19["#19–#22, #39, #67 — Done"]
    i23["#23 Cut-down LUT model — Todo, PR #92 open"]
    i24["#24 DSP56300 cycle count — Todo"]
    i106["#106 Move Octatrack work to modwerk — Todo, tracking"]
    i75["#75 v1 scope decided; shipping name still open"]
    i105["#105 Graphics and design — Todo"]

    ws10 --> i19
    ws10 --> i75
    ws10 --> i105
    ws11 --> i23
    ws11 --> i24
    ws11 --> i106

Recent decisions and merged work

Times are Europe/Amsterdam.

Time What happened Link
18 Nov, 10:21 PR #68 merged, closing #67: stereo SIMD vector maths. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/68
18 Nov, 10:46 PR #94 merged, closing #76: official x86-64 Linux stereo benchmark. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/94
18 Nov, 10:48 PR #91 merged, closing #22: first loadable AU v2 + VST 3 build. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/91
18 Nov, 21:00 PR #95 merged (refs #74): measure and test DK host sample rates. #74 later closed as completed at 22:00. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/95
18 Nov, 21:20 PR #93 merged: WDF version of the 12AX7 stage, reference only (refs #12). https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/93
18 Nov, 22:00 PR #97 merged, closing #73: bias smoothing on the audio thread. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/97
18 Nov, 22:22 PR #100 merged, closing #99: macOS CI jobs only on ready-for-review PRs. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/100
18 Nov, 22:31 PR #101 merged: AGENTS.md cross-model review process. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/101
19 Nov, 08:48 PR #98 merged, closing #96: 44.1 kHz FIR rejection and rate-aware aliasing guard. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/98
19 Nov, 17:32 Michael decided v1 scope: 6AS6 + post-distortion filter as the core, 12AX7 as an optional swap-in, filter included. Working name still open at that moment. https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/75
19 Nov, 17:42 #45 closed as completed: keep standard (8×) without pre-warp — revisit if HF accuracy matters. https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/45
19 Nov, 20:12 Michael decided the code name is Gyre. Shipping name still open; trademark search still needed. https://github.com/pmsorhaindo/ModellingAudioProcessors/issues/75
20 Nov, 00:06 PR #117 merged, closing #116: plugin identity renamed to Gyre. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/117
20 Nov, 04:13 PR #118 merged, closing #102: 6AS6 circuit values (unbypassed current-targeted cathode, P1). https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/118
20 Nov, 08:56 PR #120 merged, closing #110: bias as current-targeted cathode rheostat, validated 0.1–0.6 mA. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/120
20 Nov, 12:59 PR #119 merged: 6AS6 post-distortion low-pass, TPT SVF inside the oversampled block (refs #103). #103 remains open. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/119
20 Nov, 19:22 PR #122 merged: GitHub Action that notifies Stewart when an issue is opened. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/122
20 Nov, evening PR #121 is open (not a draft at the time of reading): T/P1/P2 mode wiring for the 6AS6 stage, with review follow-ups #123–#127. https://github.com/pmsorhaindo/ModellingAudioProcessors/pull/121

What’s next

Still waiting on Michael

Sources

Gaps