{ "schemaVersion": 1, "wave": 4, "generatedAt": "2026-07-15T19:32:16.871Z", "result": "PASS", "benchmarkSourceSha256": "02ae7b7250f6413befe31ff56b3b1a21f8f751c4ff361f7508e0ebd972beb73e", "license": { "id": "CC0-1.0", "name": "CC0 1.0 Universal", "provenance": "Synthetic, deterministic signals generated from mathematical functions by this benchmark; no recording or copyrighted composition is used." }, "environment": { "runner": "Playwright driving public tool pages", "browser": "149.0.7827.55", "server": "local Wrangler Pages" }, "methodology": [ "Generate deterministic PCM16 WAV fixtures from mathematical signals and a seeded PRNG, then record SHA-256 provenance.", "Load each public tool page through local Wrangler Pages and perform file selection, controls, processing and downloads through the real UI.", "Parse downloaded PCM16 WAVs independently in Node and measure authored frequency routing, channel relationships, reconstruction error, stationary-noise reduction and signal retention.", "Capture requests from immediately before file selection through download completion, including request type, method, origin, request-body bytes, failures and optional neural-model URLs.", "Keep every threshold fixed before the first run and retain any mismatch as FAIL rather than relaxing the criterion." ], "thresholds": { "durationSeconds": 0.003, "vocal": { "instrumentalVocalToSideMax": 0.1, "instrumentalBassToSideMin": 0.2, "sidePhaseCosineMax": -0.95, "acapellaSideToVocalMax": 0.02, "acapellaBassToVocalMax": 0.2, "acapellaChannelDifferenceMax": 0.00006103515625 }, "stems": { "reconstructionRmsMax": 0.0003, "reconstructionPeakMax": 0.002, "bassRoutingRatioMin": 5, "vocalRoutingRatioMin": 5, "vocalSideToCenterMax": 0.02, "drumSideAmplitudeMin": 0.02, "drumSidePhaseCosineMax": -0.9, "drumTransientRatioMin": 2 }, "noise": { "noiseReductionDbMin": 8, "humReductionDbMin": 10, "speechToneRetentionMin": 0.7, "snrImprovementDbMin": 6 } }, "fixtures": [ { "id": "wave4-vocal-mid-side-48k", "file": "assets/benchmark-fixtures/wave4-vocal-mid-side-48k.wav", "bytes": 576044, "sha256": "6295b8173d0be3ed68648405165f49857e5278fddac40e2620d81ffd3043f43b", "license": "CC0-1.0", "synthesis": { "kind": "stereo-mid-side-tone-matrix", "sampleRate": 48000, "channels": 2, "durationSeconds": 3, "centerComponents": [ { "frequencyHz": 80, "peakAmplitude": 0.08 }, { "frequencyHz": 1000, "peakAmplitude": 0.18 } ], "sideComponent": { "frequencyHz": 440, "peakAmplitude": 0.12, "leftPolarity": 1, "rightPolarity": -1 }, "formula": "L = center80 + center1000 + side440; R = center80 + center1000 - side440; 20 ms edge fades" } }, { "id": "wave4-stems-routing-48k", "file": "assets/benchmark-fixtures/wave4-stems-routing-48k.wav", "bytes": 768044, "sha256": "505b09b0d2b5403d51e41b8603503f1b47a41075176fdce1dc775233cc6d1509", "license": "CC0-1.0", "synthesis": { "kind": "stereo-stem-routing-matrix", "sampleRate": 48000, "channels": 2, "durationSeconds": 4, "centerComponents": [ { "frequencyHz": 80, "peakAmplitude": 0.07 }, { "frequencyHz": 1000, "peakAmplitude": 0.07 } ], "sideComponent": { "frequencyHz": 7000, "peakAmplitude": 0.045, "leftPolarity": 1, "rightPolarity": -1 }, "transients": { "frequencyHz": 800, "peakAmplitude": 0.18, "decaySeconds": 0.012, "onsetsSeconds": [ 0.5, 1, 1.5, 2, 2.5, 3, 3.5 ] }, "formula": "centred 80/1000 Hz bed + centred decaying 800 Hz bursts + anti-phase 7000 Hz side; 20 ms edge fades" } }, { "id": "wave4-noise-stationary-48k", "file": "assets/benchmark-fixtures/wave4-noise-stationary-48k.wav", "bytes": 384044, "sha256": "f29c4dee1e0c54e4fc341ca26cf3ad84f9da7b8a1b6ee62652136405cbb1b538", "license": "CC0-1.0", "synthesis": { "kind": "seeded-stationary-noise-with-speech-proxy", "sampleRate": 48000, "channels": 1, "durationSeconds": 4, "seed": "0x041A0D10", "noiseOnlySeconds": [ 0, 1 ], "stationary": { "whiteNoisePeakAmplitude": 0.03, "humFrequencyHz": 120, "humPeakAmplitude": 0.025 }, "speechProxy": [ { "frequencyHz": 500, "peakAmplitude": 0.12 }, { "frequencyHz": 1000, "peakAmplitude": 0.08 }, { "frequencyHz": 2000, "peakAmplitude": 0.04 } ], "formula": "seeded xorshift32 white noise + 120 Hz hum throughout; 500/1000/2000 Hz tones begin at 1 s" } } ], "tools": [ { "slug": "vocal-remover", "result": "PASS", "summary": "2/2 real UI instant-DSP cases routed the authored centre and side components into readable instrumental and acapella WAVs without a neural-model request.", "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0 }, "cases": [ { "id": "vocal-remover-instant-instrumental", "result": "PASS", "inputFixtures": [ "assets/benchmark-fixtures/wave4-vocal-mid-side-48k.wav" ], "outputFixture": "/assets/benchmark-results/wave4-vocal-instrumental.wav", "expected": { "engine": "instant phase DSP", "mode": "instrumental", "channels": 2, "sampleRate": 48000, "durationSeconds": 3 }, "observed": { "downloadedWavBytes": 576044, "downloadedWavSha256": "75600348cdd82d9aa9a66d779db9f6574325cb171db34a60bf89b78f27234a91", "channels": 2, "sampleRate": 48000, "durationSeconds": 3, "statusText": "✅ Ready — press play to preview, then download below.", "toneAmplitudes": { "80": 0.435959, "440": 0.531941, "1000": 0.015363 }, "sidePhaseCosine": -1, "maxChannelDifference": 1.06387329 }, "thresholds": { "instrumentalVocalToSideMax": 0.1, "instrumentalBassToSideMin": 0.2, "sidePhaseCosineMax": -0.95, "acapellaSideToVocalMax": 0.02, "acapellaBassToVocalMax": 0.2, "acapellaChannelDifferenceMax": 0.00006103515625 }, "assertions": [ { "name": "real UI instant-DSP instrumental render and WAV download completed", "pass": true, "evidence": "576044 byte WAV; ✅ Ready — press play to preview, then download below." }, { "name": "download is stereo 48 kHz PCM16 WAV", "pass": true, "evidence": "2 channel(s), 48000 Hz, 16-bit format 1" }, { "name": "duration is preserved", "pass": true, "evidence": "expected 3 s, observed 3 s" }, { "name": "centred 1 kHz vocal is strongly reduced relative to the side signal", "pass": true, "evidence": "1 kHz / 440 Hz = 0.02888" }, { "name": "anti-phase 440 Hz side signal remains anti-phase", "pass": true, "evidence": "phase cosine -1" }, { "name": "low centred bass foundation remains measurable", "pass": true, "evidence": "80 Hz / 440 Hz = 0.819562" }, { "name": "no workflow network request", "pass": true, "evidence": "0 request(s) after file selection" }, { "name": "no payload, cross-origin or request-body transfer", "pass": true, "evidence": "0 payload, 0 cross-origin, 0 request-body byte(s)" }, { "name": "no non-GET workflow request", "pass": true, "evidence": "0 non-GET request(s)" }, { "name": "no optional neural runtime or model request", "pass": true, "evidence": "0 neural runtime/model request(s)" }, { "name": "no console or page error", "pass": true, "evidence": "0 console error(s), 0 page error(s)" } ], "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0, "requests": [] }, "limitations": [ "This case tests the instant phase-DSP engine only. It does not load or score the optional neural model.", "Output peak normalization means routing is assessed with component ratios and channel relationships, not absolute gain preservation." ] }, { "id": "vocal-remover-instant-acapella", "result": "PASS", "inputFixtures": [ "assets/benchmark-fixtures/wave4-vocal-mid-side-48k.wav" ], "outputFixture": "/assets/benchmark-results/wave4-vocal-acapella.wav", "expected": { "engine": "instant phase DSP", "mode": "acapella", "channels": 2, "sampleRate": 48000, "durationSeconds": 3 }, "observed": { "downloadedWavBytes": 576044, "downloadedWavSha256": "41715eb19bbe696b3d12763da206e0a5c826e97d97d6cdb8809a260599818b92", "channels": 2, "sampleRate": 48000, "durationSeconds": 3, "statusText": "✅ Ready — press play to preview, then download below.", "toneAmplitudes": { "80": 0.085425, "440": 0.000002, "1000": 0.894699 }, "sidePhaseCosine": 1, "maxChannelDifference": 0 }, "thresholds": { "instrumentalVocalToSideMax": 0.1, "instrumentalBassToSideMin": 0.2, "sidePhaseCosineMax": -0.95, "acapellaSideToVocalMax": 0.02, "acapellaBassToVocalMax": 0.2, "acapellaChannelDifferenceMax": 0.00006103515625 }, "assertions": [ { "name": "real UI instant-DSP acapella render and WAV download completed", "pass": true, "evidence": "576044 byte WAV; ✅ Ready — press play to preview, then download below." }, { "name": "download is stereo 48 kHz PCM16 WAV", "pass": true, "evidence": "2 channel(s), 48000 Hz, 16-bit format 1" }, { "name": "duration is preserved", "pass": true, "evidence": "expected 3 s, observed 3 s" }, { "name": "anti-phase 440 Hz side signal is rejected from the acapella", "pass": true, "evidence": "440 Hz / 1 kHz = 0.000002" }, { "name": "80 Hz bass is attenuated relative to the vocal band", "pass": true, "evidence": "80 Hz / 1 kHz = 0.095479" }, { "name": "acapella channels are sample-identical within PCM16 rounding", "pass": true, "evidence": "max channel difference 0" }, { "name": "no workflow network request", "pass": true, "evidence": "0 request(s) after file selection" }, { "name": "no payload, cross-origin or request-body transfer", "pass": true, "evidence": "0 payload, 0 cross-origin, 0 request-body byte(s)" }, { "name": "no non-GET workflow request", "pass": true, "evidence": "0 non-GET request(s)" }, { "name": "no optional neural runtime or model request", "pass": true, "evidence": "0 neural runtime/model request(s)" }, { "name": "no console or page error", "pass": true, "evidence": "0 console error(s), 0 page error(s)" } ], "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0, "requests": [] }, "limitations": [ "This case tests the instant phase-DSP engine only. It does not load or score the optional neural model.", "Output peak normalization means routing is assessed with component ratios and channel relationships, not absolute gain preservation." ] } ] }, { "slug": "stem-splitter", "result": "PASS", "summary": "1/1 real UI workflow exported four routed stems whose independently parsed sum reconstructed the input within the published error bounds.", "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0 }, "cases": [ { "id": "stem-splitter-four-export-reconstruction", "result": "PASS", "inputFixtures": [ "assets/benchmark-fixtures/wave4-stems-routing-48k.wav" ], "outputFixtures": { "bass": "/assets/benchmark-results/wave4-stem-bass.wav", "vocals": "/assets/benchmark-results/wave4-stem-vocals.wav", "drums": "/assets/benchmark-results/wave4-stem-drums.wav", "other": "/assets/benchmark-results/wave4-stem-other.wav" }, "expected": { "stemIds": [ "bass", "vocals", "drums", "other" ], "channels": 2, "sampleRate": 48000, "durationSeconds": 4 }, "observed": { "downloadedWavBytes": { "bass": 768044, "vocals": 768044, "drums": 768044, "other": 768044 }, "downloadedWavSha256": { "bass": "9095163b4d9a869d768de8b5a755792c980b9039de2bb9fb4f94069ffc14cf70", "vocals": "6019e54edfe8d8da068b844c549fcbd96d429e8d4de9e58983105e43b48938cd", "drums": "8402de4cec3f6d059de69a4c74b9711d6c14e89e9b5ec302cedcdf804d82e3ae", "other": "3a8dc115b0418b2324c755949ce21de1c8b4e095b8981811f2808a4bb50a37e0" }, "statusText": "✅ All 4 stems ready — scroll down to preview and download.", "reconstruction": { "rms": 0.00003146, "peak": 0.00009155 }, "routing": { "bass80Ratio": 51.155421, "vocal1000Ratio": 5.887089, "vocalSideRatio": 0.000067, "drum7000Amplitude": 0.055657, "drumSidePhaseCosine": -1, "drumTransientRatio": 62.645702 } }, "thresholds": { "reconstructionRmsMax": 0.0003, "reconstructionPeakMax": 0.002, "bassRoutingRatioMin": 5, "vocalRoutingRatioMin": 5, "vocalSideToCenterMax": 0.02, "drumSideAmplitudeMin": 0.02, "drumSidePhaseCosineMax": -0.9, "drumTransientRatioMin": 2 }, "assertions": [ { "name": "real UI split and four WAV downloads completed", "pass": true, "evidence": "4/4 WAVs; ✅ All 4 stems ready — scroll down to preview and download." }, { "name": "all stems are stereo 48 kHz PCM16 WAVs with preserved duration", "pass": true, "evidence": "2ch/48000Hz/16bit/4s, 2ch/48000Hz/16bit/4s, 2ch/48000Hz/16bit/4s, 2ch/48000Hz/16bit/4s" }, { "name": "four downloaded stems reconstruct the input within fixed error bounds", "pass": true, "evidence": "RMSE 0.00003146, peak 0.00009155" }, { "name": "bass stem routes authored 80 Hz above mid/high components", "pass": true, "evidence": "80 Hz routing ratio 51.1554" }, { "name": "vocal stem routes centred 1 kHz above 80 Hz", "pass": true, "evidence": "1 kHz / 80 Hz = 5.8871" }, { "name": "vocal stem rejects the anti-phase 7 kHz side component", "pass": true, "evidence": "7 kHz / 1 kHz = 0.000067" }, { "name": "drum stem retains anti-phase high-side energy", "pass": true, "evidence": "7 kHz amplitude 0.055657, phase cosine -1" }, { "name": "drum transient gate raises centred burst energy above inter-burst energy", "pass": true, "evidence": "mid burst/quiet RMS ratio 62.6457" }, { "name": "no workflow network request", "pass": true, "evidence": "0 request(s) after file selection" }, { "name": "no payload, cross-origin or request-body transfer", "pass": true, "evidence": "0 payload, 0 cross-origin, 0 request-body byte(s)" }, { "name": "no non-GET workflow request", "pass": true, "evidence": "0 non-GET request(s)" }, { "name": "no console or page error", "pass": true, "evidence": "0 console error(s), 0 page error(s)" } ], "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0, "requests": [] }, "limitations": [ "This tests the deterministic frequency/transient heuristic and sample-level reconstruction, not semantic or AI-grade stem separation.", "Filter responses may vary slightly across browser engines; fixed routing ratios intentionally avoid exact filtered-sample expectations." ] } ] }, { "slug": "noise-remover", "result": "PASS", "summary": "1/1 real UI cleanup reduced seeded stationary noise and hum while retaining the authored speech-proxy tones and improving SNR.", "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0 }, "cases": [ { "id": "noise-remover-stationary-profile-default", "result": "PASS", "inputFixtures": [ "assets/benchmark-fixtures/wave4-noise-stationary-48k.wav" ], "outputFixture": "/assets/benchmark-results/wave4-noise-denoised.wav", "expected": { "strength": 1, "channels": 1, "sampleRate": 48000, "durationSeconds": 4 }, "observed": { "downloadedWavBytes": 384044, "downloadedWavSha256": "5201344ccabaf8141c2be119c5ab804bfbf699c91fe5c5c971e29fd4a04589be", "statusText": "✅ Cleaned — A/B it against the original above, then download.", "noiseOnlyRms": { "input": 0.02459722, "output": 0.00023526, "reductionDb": 40.38673 }, "hum120Hz": { "inputAmplitude": 0.02480079, "outputAmplitude": 2.1e-7, "reductionDb": 101.365476 }, "speechProxy": { "inputToneAmplitudeSum": 0.239723, "outputToneAmplitudeSum": 0.226695, "retention": 0.945657, "inputSnrDb": 12.575549, "outputSnrDb": 35.633261, "improvementDb": 23.057712 } }, "thresholds": { "noiseReductionDbMin": 8, "humReductionDbMin": 10, "speechToneRetentionMin": 0.7, "snrImprovementDbMin": 6 }, "assertions": [ { "name": "real UI default-strength cleanup and WAV download completed", "pass": true, "evidence": "384044 byte WAV; ✅ Cleaned — A/B it against the original above, then download." }, { "name": "download is mono 48 kHz PCM16 WAV with preserved duration", "pass": true, "evidence": "1ch/48000Hz/16bit/4s" }, { "name": "noise-only RMS is reduced by at least 8 dB", "pass": true, "evidence": "40.3867 dB" }, { "name": "authored 120 Hz stationary hum is reduced by at least 10 dB", "pass": true, "evidence": "101.3655 dB" }, { "name": "speech-proxy tones retain at least 70% of their input amplitude", "pass": true, "evidence": "94.566% retained" }, { "name": "speech-proxy SNR improves by at least 6 dB", "pass": true, "evidence": "23.0577 dB improvement" }, { "name": "no workflow network request", "pass": true, "evidence": "0 request(s) after file selection" }, { "name": "no payload, cross-origin or request-body transfer", "pass": true, "evidence": "0 payload, 0 cross-origin, 0 request-body byte(s)" }, { "name": "no non-GET workflow request", "pass": true, "evidence": "0 non-GET request(s)" }, { "name": "no console or page error", "pass": true, "evidence": "0 console error(s), 0 page error(s)" } ], "networkDataFlow": { "requestsDuringWorkflow": 0, "payloadRequests": 0, "crossOriginRequests": 0, "nonGetRequests": 0, "requestBodyBytes": 0, "modelRequests": 0, "badResponses": 0, "failedRequests": 0, "requests": [] }, "limitations": [ "This case covers deterministic stationary broadband noise and hum, not changing noise, reverb, overlapping voices or isolated transient sounds.", "Measurement windows exclude start/end and the one-second transition so the benchmark does not score overlap-add edge handling as denoising quality." ] } ] } ], "limitations": [ "Synthetic fixtures isolate deterministic behavior; they do not establish perceptual quality for mastered music, human speech or noisy field recordings.", "The vocal-remover cases cover only the instant phase-DSP engine. The optional neural engine, external runtime and model download are explicitly outside this wave.", "Stem labels describe frequency/transient heuristics. Reconstruction and routing evidence must not be presented as AI-grade semantic separation.", "The denoising case covers one stationary seeded profile. Changing noise, reverberation, overlapping voices and one-off sounds remain outside this wave.", "Web Audio filter responses can vary slightly between browser engines; the fixed thresholds use broad ratios rather than exact filtered samples." ], "fatalError": null }