"CultMesh speech render/scoring orchestration with payload manifests, peer/lease admission receipts, worker assignment, result collection, and central checkpoint application while the authority trainer owns optimizer steps and checkpoint commits",
"local song-snippet scene parity challenges with seeded clip extraction, tempo autocorrelation, register/scale hints, external Codex swarm orchestration, rendered candidate scoring, and song trial-result CultCache records",
"status":"This is now the active project center. Roadmap and research notes align the curriculum around structured utterance metadata -> learned utterance embedding -> learned synth automation -> vocal tract/envelope/LFO/filter/spectral controls -> rendered audio loss. The first Core contracts still defend the boundary: PhoneticIntent is the Weksa/language-owned handoff, ArticulatoryPlan is the AquaSynth realization layer, VocalTractMorphology declares anatomy capabilities, ArticulatoryConstraintReport is the host-visible veto surface, and VocalTractPlanResult is the tiny end-to-end proof artifact for host branching. IPA is an authoring input only: tokenization and feature mapping lower into language phonology profiles, timed articulatory gestures, morphology-specific tract/excitation curves, and Faust DSP. The staged support plan starts with tokenization and vowels, then core pulmonic consonants, expanded human IPA, airstream/phonation, and Weksa/alien morphologies. Optional eSpeak NG tests provide the first generated speech workout when ESPEAK_NG or espeak-ng is installed; eSpeak is intelligibility/timing/log-mel pressure, not anatomy truth. AquaSynth owns the learned utterance embedding encoder and synth driver: speech text embeddings plus prosody/emphasis hints plus Ghostlight/Epiphany-shaped character state vectors compress into utterance embeddings, and phonetic features plus that embedding feed a packed-buffer native C# backprop/Adam mapper that predicts tract, AM/FM, LFO, filter, and mel spectral-envelope controls. SpeechBackpropagationPipeline chains the two models: synth-driver target loss returns gradients with respect to its input, slices the utterance-embedding gradient, and updates the utterance encoder from downstream speech loss. The generated ground-truth training fixture renders eSpeak syllables, extracts log-mel evidence, and trains that chained pipeline toward eSpeak-seeded mel/control targets. CultMesh speech training now has payload manifests, peer/lease admission receipts, worker assignment, result collection, and central checkpoint application; workers do not own model updates. Real remote artifact chunk transfer and process supervision remain CultMesh infrastructure surfaces, not AquaSynth workaround territory. IpaGestureExperiment writes frozen descriptor/variant rounds with candidate scripts, primitive timelines, gesture metrics, manifests, and evidence JSONL, then analyzes them into metric summaries, score-band clusters, and a science brief. IpaTrialOrchestrator now writes local five-seed IPA render trials, accepts externally authored candidate scripts through RunCandidateScriptsAsync, writes WAV/report artifacts, evaluator reports, extracted primitive timeline facts, and `ipa-trial-results.cc` CultCache records. `tools/IpaTrialWorker` is the public worker tool for seed/score/search/show/index over the store; search now auto-upserts Qdrant/Ollama evidence chunks including timeline-fact text and returns vector-backed reports, while `tools/run-ipa-trial-loop.ps1` dispatches external Codex hypothesis/evaluator workers around five five-phoneme target sets. The latest seed run improved voiced vowels/nasals after primitive `SourcePort` lowering gained an oscillatory carrier; plosive closure/release remains the next weak owner.",
"lesson":"Do not build a complete IPA parser before the tract can say basic syllables. The durable machine is tokenizer -> feature mapper -> phonology profile -> gesture planner -> morphology -> tract plan -> Faust renderer, with dependency-injected seams and mocks at each boundary. Weksa should use IPA-like phonetic input without letting human IPA semantics own alien anatomy. Anatomy has veto power: if a morphology lacks bilabial closure or another required capability, AquaSynth reports that failure instead of smearing over it.",
"next":"Start the next speech gradient descent pass from Weksa's emitted v0.1 seed batch at `E:\\Projects\\weksa\\examples\\speech-training\\tiny-panphon-v0.1\\batch.json`: 1024 bge-m3 semantic speech-text floats, variable-length PanPhon sequence evidence, 256 AquaSynth-owned phonetic-realization embedding floats from `aquasynth.panphon_sequence_encoder.v0.1`, 32 prosody/emphasis floats, 64 projected character-state floats, and 64 AquaSynth utterance-embedding output floats. Keep semantic text embeddings and phonetic realization separate. Next tighten the parity harness: load that batch into AquaSynth training records, render tiny utterance fixtures, sweep controller outputs through compiled Faust controls, score log-mel/timing/envelope loss, record per-stage decision time/latency/budget/confidence receipts, backpropagate into the utterance encoder plus synth driver, and connect Stage 0/1 IPA token spans, vowel feature bundles, and human baseline tract-area targets to that loop. For distributed work, use RunCultMeshTrainingStep as the boring authority path: admitted leased workers return gradients/results, then the trainer applies centrally. Do not attempt federated averaging until this path is exercised against real Faust-rendered speech jobs."
"status":"DX7 SysEx parser extracts voice edit buffers, packed voices, packed 32-voice banks, six operators, envelopes, algorithm, feedback, oscillator sync, LFO, transpose, voice name, checksum validation, ReferencePatch features, algorithm topology from the reverse-engineered OPS algorithm ROM table, envelope-to-ADSR approximations, measured carrier output amplitude, isolated two-op modulation route scale, summed/cascaded route scale, calibrated feedback amount table, ratio detune lowering, voice transpose lowering, fixed-frequency operator lowering, ROM COM output compensation cut, operator key/rate scaling at the effective played note, topology-aware route lowering, staged operator envelopes with per-segment curves, raw and block-interpolated DX7 envelope tracing, applied-envelope curved lowering with optional sustain floor, two AquaSynth DSL rebuild attempts, first-class OperatorGraph support with graph-level vibrato, public-domain SysEx fixture coverage, dexed-py reference rendering, project-authored Dexed patch rendering, AquaSynth candidate rendering through Faust-generated C#, one thresholded rendered parity test for analog1.syx voice 13, a hard PRC SYNTH1 parity probe with listening WAV artifacts and log-mel <= 0.145, project-authored algorithm-8 summed-stack/cascade/summed-pair parity, rendered community parity for Mooger, Piano Bass, RES SYNTH1, and ANLGSYN 1 with a sustained high-band gate, a MELLOWSOLO pressure artifact test, and renderable delayed operator feedback.",
"lesson":"Algorithm 32's additive six-carrier shape maps cleanly to six AquaSynth voices. Algorithm 8 justified OperatorGraph: operators, modulation edges, carriers, feedback, and graph-level pitch motion belong to a graph authority, not Voice fields. The first rendered parity rung is behavioral and sine-like, not exact DX7 operator execution. A harder algorithm-8 target exposed that feedback must be delayed recursion in Faust, not a direct cyclic self-reference. Static DX7 output-level approximation helped import/rebuild math but was mathematically wrong for carrier amplitude. A single-carrier Dexed sweep shows carrier amplitude follows 2^((outputLevel - 99) / 8). Isolated two-op DX7 FM maps full-level modulation to AquaSynth route index 6.275; summed/cascaded algorithm-8 branches now use 6.0 after PRC listening, a project-authored combined stack probe, and split cascade/summed-pair probes showed lower values were under-bright. Feedback lowers through a fitted table instead of PRC-local constants; ANLGSYN 1 showed max feedback value 7 needs a hotter AquaSynth amount 2.2 to recover buzzy high bands, while value 5 remains 0.19 for PRC. ANLGSYN 1 then exposed a sustained-buzz gap: output soft clip helped but did not stop mid-note drift; max-feedback source operators need an applied-envelope sustain floor so the feedback loop stays driven after the attack. ANLGSYN 1's remaining beat is pitch LFO: AMD is nonzero but all operator AMS values are 0, while PMD/PMS lower cleanly to operator-graph vibrato with faded delay. Ratio-mode detune belongs in operator frequency lowering. DX7 voice transpose belongs in graph base frequency: transpose 24 is neutral, transpose 12 lowers MIDI note 60 to 130.8128 Hz. DX7 fixed-frequency operators are absolute Hz, not note ratios; ANLGSYN 1 proved fixed op1/op3 must lower to ratios around 0.007615/0.009227 against a 130.8128 Hz graph base to preserve low buzzing modulation. ROM COM carrier compensation was cut: user listening found no-COM renders harmonically closer and mainly too quiet, so `OperatorOutputCompensation` returns unity and graph gain owns loudness recovery. Operator key-scaling and rate-scaling must lower against the effective played MIDI note after transpose; hardcoded MIDI 60 made RES SYNTH1 too soft on attack and Piano Bass less harmonically accurate. Project-authored Dexed patch rendering now gives lawful calibration targets. Reusing the operator output curve for EG levels was tested and cut because it failed hard-target log-mel; later output-level-dependent and naive traced-release applied-envelope experiments were also cut because they worsened broader parity. Direct envelope tracing shows current linear `env=rl` cannot express DX7 EG spike/drop behavior. Dexed-style audio samples EG once per 64-sample block and interpolates applied gain, so raw EG parity alone is not audio parity. Staged operator envelopes now support linear/exponential segments and supranormal levels. `ApproximateAppliedRateLevelEnvelope` uses that surface and now picks the first near-peak before gate so slower rising modulators hit early instead of swelling late; it can also preserve a floor for max-feedback source sustain. Community gains with no COM and played-note scaling: Mooger 0.75, Piano Bass 0.72, RES SYNTH1 0.75. Algorithm-2 non-carrier applied envelopes currently cap the first peak at 0.92 to prevent ANLGSYN 1/RES SYNTH1 peak blowout while preserving Piano Bass. MELLOWSOLO is a pressure artifact test, not passing parity; its render window is 1.25s so the candidate release is audible instead of cut at the artifact boundary. Mooger needs a more aggressive attack contour, and RES SYNTH1's phase-like attack issue is likely relative operator attack/peak shape because oscillator sync is enabled. PRC remains passing on log-mel/envelope/RMS, with score floor lowered to 0.74 after rate scaling changed the aggregate score.",
"next":"Use the now-validated algorithm-8 route scale, transpose lowering, and fixed-frequency lowering as fixed inputs while isolating remaining community-voice timbre mismatch, then expand to broader public-domain banks."
"status":"Initial ZynInstrumentReader parses plain XML and gzip-compressed .xiz-shaped XML from project-authored fixtures into ZynInstrument records and ReferencePatch features. First Zyn-style AquaSynth rebuilds exist for additive lead, PAD texture, and vocal/formant layer. `ZynInstrumentSurvey` can rank .xiz directories by feature pressure. Named `layer` declarations provide the first Zyn-driven DSL abstraction; additive `harmonics`, PAD `spectrum`, and staged `env=rl` voice envelopes now attach to those layers. Zyn PAD `spectrum` banks can now carry explicit `zyn_mode`, `zyn_bandwidth`, `zyn_bwscale`, `zyn_profile`, and `zyn_position` table-generation fields, and `PadSynthWaveform` lowers those through Zyn-style profile/bandwidth math. The pinned Zyn source now builds as a test-only MSYS2/MinGW oracle through `tools/zyn-reference/build-zyn-reference.ps1`, which patches an ignored artifact source copy, proves `PadNoteTest`, and builds `ZynPadReference`.",
"next":"Continue Zyn PAD parity from the sampled OscilGen stage-order, multi-kit PAD, Zyn-curve static global-filter, low-pass/high-pass order, PAD frequency-detune, low-pass/high-pass filter-envelope, logarithmic Zyn envelope-time, generated Zyn PAD output-calibration, low-pass note tracking, explicit low-pass Q, Q+order cascade, one-based OscilGen oscillator-filter index, base-function-7 selected-table-root, base-function phase modulation, and adaptive-harmonics paths. Ghost Ensemble now expands its power-base multi-harmonic OscilGen source, applies Zyn filter note tracking from `freq_track`, emits `lpf_q` from the Zyn analog filter `q`, and preserves low-pass slope by cascading resonant Faust stages according to `lpf_order`; latest Ghost log-mel is 0.369402, envelope distance is 0.363685, RMS ratio is 1.054875, centroid ratio is 1.006808, and score is 0.57229. DoublePadBass is accepted by ear and latest survey reports log-mel 0.247674, envelope distance 0.208351, RMS ratio 0.888011, score 0.650319. Soft Pad is accepted by ear. Organ Choir now maps kit0 HP2/stage/filter-envelope motion and kit1's selected table root, `base_function_modulation=1`, and `adaptive_harmonics=1`; latest Organ log-mel is 0.471483, envelope distance 0.302978, RMS ratio 0.878, centroid ratio 1.011328, and score 0.541933. Its kit1 PAD table energy now matches the wind-body shape much more closely: 400-800 Hz 0.2713 vs Zyn 0.2780 and 8-16 kHz 0.1471 vs Zyn 0.1355. ZynPadReference has test-only `harmonics` and `selected` modes; selected roots are diagnostic/source-specific, not a global default, because global selected roots regressed accepted DoublePadBass. sin2x remains loud/concentrated after the one-based filter-index fix, so do not claim full Zyn PAD parity until broader OscilGen filtering/waveshaping/adaptive-harmonic behavior and note/global filter LFO ownership are mapped or deliberately excluded."
"sfxr_mapping":"SFXR sustain is note gate duration. SFXR punch maps to lower ADSR sustain level plus voice gain compensation so the implicit attack peak stays at 1 and the held level sits below it.",
"faust_surface":"One-shot patches use generated ADSR over patch age and note gate duration. MIDI mono/poly patches emit Faust's standard freq/gain/gate controls and [midi:on][nvoices:n] metadata so Faust architectures own note allocation."
"aquasynth_requirement":"AquaSynth DSL patches declare stable exposed parameters with path/default/min/max/step metadata so callers can vary sound without recompiling.",
"binding_rule":"Parameter binding uses field-site references such as `lpf=@/macro/brightness` and `/opgraphs/0/operators/2/level`; the field owns meaning and the parameter owns the runtime value. Avoid target-routing semantics unless a new graph authority is deliberately mapped."
"result":"Three-phase song-snippet experiment across different seeded segments completed under artifacts/parity/song-snippet-swarms/20260531T014033022. Best log-mel cosine improved to 0.380805 through formant/noise scene roles, still pressure rather than close parity.",
