Add bioacoustic training receipts

This commit is contained in:
Meta Vens 2026-05-24 11:41:04 +01:00
commit 84d33992e8
91 changed files with 2416 additions and 44 deletions

View file

@ -0,0 +1,84 @@
{
"ResultId": "bioacoustic-20260524-103438:baseline-mfcc-index:blur-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "baseline-mfcc-index",
"HypothesisNotes": "Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
"DegradationName": "blur-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9817037980063056,
"TimingMeanAbsoluteErrorSamples": 1712,
"DecodeMilliseconds": 151.211,
"RealtimeFactor": 3.3066377446085267,
"IdentityScore": 0.9981703798006306,
"TimingScore": 0.12295081967213116,
"SpeedScore": 0.06613275489217053,
"TotalScore": 0.6542233339958087,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0007356480619029866,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "baseline-mfcc-index/blur-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "baseline-mfcc-index/blur-light/post-warp.wav",
"ContentHash": "46984f1c8f4b37f309173dc8926769b24a95403f15b279da9d9fb1e6a1f971be"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "baseline-mfcc-index/blur-light/reconstructed-from-detections.wav",
"ContentHash": "5c4d4cda22373af4034310c85510c724cfe807d11de6fc094ffb09414813c7c3"
},
{
"Kind": "summary-json",
"Uri": "baseline-mfcc-index/blur-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1920,
"Confidence": 0.9842773002273854
},
{
"EventIndex": 1,
"SampleOffset": 9600,
"Confidence": 0.9759566210616761
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9848774727298553
}
]
}

View file

@ -0,0 +1,84 @@
{
"ResultId": "bioacoustic-20260524-103438:baseline-mfcc-index:clean-roundtrip",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "baseline-mfcc-index",
"HypothesisNotes": "Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
"DegradationName": "clean-roundtrip",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9711336256626195,
"TimingMeanAbsoluteErrorSamples": 432,
"DecodeMilliseconds": 162.5302,
"RealtimeFactor": 3.0763513488570124,
"IdentityScore": 0.997113362566262,
"TimingScore": 0.35714285714285715,
"SpeedScore": 0.06152702697714025,
"TotalScore": 0.6948014044722247,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.008886063472396654,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "baseline-mfcc-index/clean-roundtrip/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "baseline-mfcc-index/clean-roundtrip/post-warp.wav",
"ContentHash": "f803fceaf796518f32920b9c777b27e7017379fcdbc6acb46e2c89889ef1aa74"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "baseline-mfcc-index/clean-roundtrip/reconstructed-from-detections.wav",
"ContentHash": "06c93a4247b73e9ed46e7d18737924aa0923baa3f3ffe10ccb3a216684a93173"
},
{
"Kind": "summary-json",
"Uri": "baseline-mfcc-index/clean-roundtrip/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 3840,
"Confidence": 0.9681068418929425
},
{
"EventIndex": 1,
"SampleOffset": 11520,
"Confidence": 0.9751340322412796
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9701600028536365
}
]
}

View file

@ -0,0 +1,84 @@
{
"ResultId": "bioacoustic-20260524-103438:baseline-mfcc-index:warp-heavy-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "baseline-mfcc-index",
"HypothesisNotes": "Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
"DegradationName": "warp-heavy-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 2,
"Precision": 0.6666666666666666,
"Recall": 0.6666666666666666,
"Confidence": 0.9146826945312526,
"TimingMeanAbsoluteErrorSamples": 6831.999999999999,
"DecodeMilliseconds": 200.7895,
"RealtimeFactor": 2.490170053712968,
"IdentityScore": 0.6914682694531252,
"TimingScore": 0.03393665158371041,
"SpeedScore": 0.04980340107425936,
"TotalScore": 0.44477960456085736,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.00033629216572444076,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "baseline-mfcc-index/warp-heavy-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "baseline-mfcc-index/warp-heavy-blur/post-warp.wav",
"ContentHash": "f9a185f36c36e210bd3673e625234a7ced9bfb7b199b08d83e11d4ec8ccd93e9"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "baseline-mfcc-index/warp-heavy-blur/reconstructed-from-detections.wav",
"ContentHash": "7623c69fcd283397293985122b0c55baeab03173cb128733a8a02a3ab837d66e"
},
{
"Kind": "summary-json",
"Uri": "baseline-mfcc-index/warp-heavy-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 0,
"Confidence": 0.9501302695423681
},
{
"EventIndex": 3,
"SampleOffset": 11520,
"Confidence": 0.8335026537145045
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9604151603368851
}
]
}

View file

@ -0,0 +1,89 @@
{
"ResultId": "bioacoustic-20260524-103438:baseline-mfcc-index:warp-light-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "baseline-mfcc-index",
"HypothesisNotes": "Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
"DegradationName": "warp-light-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.9031864011599247,
"TimingMeanAbsoluteErrorSamples": 15684.000000000002,
"DecodeMilliseconds": 223.0421,
"RealtimeFactor": 2.2417292520111674,
"IdentityScore": 0.8778186401159925,
"TimingScore": 0.015071590052750564,
"SpeedScore": 0.04483458504022335,
"TotalScore": 0.5559273600894551,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0003236487418848957,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "baseline-mfcc-index/warp-light-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "baseline-mfcc-index/warp-light-blur/post-warp.wav",
"ContentHash": "a7747c66b5bd084995a439c886b9867bf2578532f204d625e03772f58a31b16e"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "baseline-mfcc-index/warp-light-blur/reconstructed-from-detections.wav",
"ContentHash": "622fe5efa1313fc63c95c322ddf08ab68e15a2011d7147d8c24164ed3bc35da8"
},
{
"Kind": "summary-json",
"Uri": "baseline-mfcc-index/warp-light-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1920,
"Confidence": 0.9562809085213103
},
{
"EventIndex": 8,
"SampleOffset": 7680,
"Confidence": 0.7797562610548343
},
{
"EventIndex": 1,
"SampleOffset": 9600,
"Confidence": 0.905395380105711
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.971313054957843
}
]
}

View file

@ -0,0 +1,89 @@
{
"ResultId": "bioacoustic-20260524-103438:baseline-mfcc-index:warp-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "baseline-mfcc-index",
"HypothesisNotes": "Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
"DegradationName": "warp-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.8964671480715712,
"TimingMeanAbsoluteErrorSamples": 5123.999999999999,
"DecodeMilliseconds": 182.7798,
"RealtimeFactor": 2.7355320445694766,
"IdentityScore": 0.8771467148071572,
"TimingScore": 0.0447427293064877,
"SpeedScore": 0.05471064089138953,
"TotalScore": 0.5628267826338831,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 9.12669517854302E-07,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "baseline-mfcc-index/warp-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "baseline-mfcc-index/warp-light/post-warp.wav",
"ContentHash": "fc695aa2e0579eab784d91523d9162deb58e036333ee6f19c185bdb90a449630"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "baseline-mfcc-index/warp-light/reconstructed-from-detections.wav",
"ContentHash": "ab3de6b3d1062ba8cb5bb47a3dbc0fbcb7b888155bb55ba26c7cb1033e4da928"
},
{
"Kind": "summary-json",
"Uri": "baseline-mfcc-index/warp-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 3840,
"Confidence": 0.9304741075993451
},
{
"EventIndex": 3,
"SampleOffset": 13440,
"Confidence": 0.8292640500764266
},
{
"EventIndex": 1,
"SampleOffset": 17280,
"Confidence": 0.8973217506582013
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9288086839523123
}
]
}

View file

@ -0,0 +1,83 @@
{
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:blur-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "compact-fast-index",
"HypothesisNotes": "Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
"DegradationName": "blur-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9821983546360261,
"TimingMeanAbsoluteErrorSamples": 2352,
"DecodeMilliseconds": 113.8517,
"RealtimeFactor": 4.3916779459595245,
"IdentityScore": 0.9982198354636026,
"TimingScore": 0.09259259259259259,
"SpeedScore": 0.0878335589191905,
"TotalScore": 0.6531296764379384,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0007356480619029866,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 32,
"CepstralCoefficients": 10,
"TableCount": 3,
"HashBits": 12,
"NearHashRadius": 3,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 6,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "compact-fast-index/blur-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "compact-fast-index/blur-light/post-warp.wav",
"ContentHash": "46984f1c8f4b37f309173dc8926769b24a95403f15b279da9d9fb1e6a1f971be"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "compact-fast-index/blur-light/reconstructed-from-detections.wav",
"ContentHash": "afb10e7ff19086f72fb86e3a59b8eb3252654707080845ec5cc906808462f978"
},
{
"Kind": "summary-json",
"Uri": "compact-fast-index/blur-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 7680,
"Confidence": 0.9832276994197949
},
{
"EventIndex": 1,
"SampleOffset": 9600,
"Confidence": 0.976671497123334
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9866958673649495
}
]
}

View file

@ -0,0 +1,83 @@
{
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:clean-roundtrip",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "compact-fast-index",
"HypothesisNotes": "Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
"DegradationName": "clean-roundtrip",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9734685897698672,
"TimingMeanAbsoluteErrorSamples": 432,
"DecodeMilliseconds": 118.0861,
"RealtimeFactor": 4.234198605932451,
"IdentityScore": 0.9973468589769867,
"TimingScore": 0.35714285714285715,
"SpeedScore": 0.08468397211864902,
"TotalScore": 0.6995775612751758,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.008886063472396654,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 32,
"CepstralCoefficients": 10,
"TableCount": 3,
"HashBits": 12,
"NearHashRadius": 3,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 6,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "compact-fast-index/clean-roundtrip/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "compact-fast-index/clean-roundtrip/post-warp.wav",
"ContentHash": "f803fceaf796518f32920b9c777b27e7017379fcdbc6acb46e2c89889ef1aa74"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "compact-fast-index/clean-roundtrip/reconstructed-from-detections.wav",
"ContentHash": "315a1e8cc8b7d35e659a692afc9a96e15a35e1fba859b57a2f147f477f560a9b"
},
{
"Kind": "summary-json",
"Uri": "compact-fast-index/clean-roundtrip/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 3840,
"Confidence": 0.9697246459110085
},
{
"EventIndex": 1,
"SampleOffset": 11520,
"Confidence": 0.9778715287480338
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9728095946505593
}
]
}

View file

@ -0,0 +1,98 @@
{
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:warp-heavy-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "compact-fast-index",
"HypothesisNotes": "Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
"DegradationName": "warp-heavy-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 6,
"CorrectAnchors": 2,
"Precision": 0.3333333333333333,
"Recall": 0.6666666666666666,
"Confidence": 0.642324401027618,
"TimingMeanAbsoluteErrorSamples": 232856,
"DecodeMilliseconds": 99.7631,
"RealtimeFactor": 5.0118731274389035,
"IdentityScore": 0.5142324401027618,
"TimingScore": 0.0010296186978755535,
"SpeedScore": 0.10023746254877806,
"TotalScore": 0.33905693673908555,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.00033629216572444076,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 32,
"CepstralCoefficients": 10,
"TableCount": 3,
"HashBits": 12,
"NearHashRadius": 3,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 6,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "compact-fast-index/warp-heavy-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "compact-fast-index/warp-heavy-blur/post-warp.wav",
"ContentHash": "f9a185f36c36e210bd3673e625234a7ced9bfb7b199b08d83e11d4ec8ccd93e9"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "compact-fast-index/warp-heavy-blur/reconstructed-from-detections.wav",
"ContentHash": "e2c15c0fdbbc805e3eba515fdcde5921dc44f604e4b91b789c4d0b4d9c0a6780"
},
{
"Kind": "summary-json",
"Uri": "compact-fast-index/warp-heavy-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 0,
"Confidence": 0.9551977551518811
},
{
"EventIndex": 13,
"SampleOffset": 7680,
"Confidence": 0.6949601234449054
},
{
"EventIndex": 32,
"SampleOffset": 9600,
"Confidence": 0.5239472197964249
},
{
"EventIndex": 55,
"SampleOffset": 11520,
"Confidence": 0.39955705873729075
},
{
"EventIndex": 85,
"SampleOffset": 15360,
"Confidence": 0.3168909673131934
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9633932817220121
}
]
}

View file

@ -0,0 +1,93 @@
{
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:warp-light-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "compact-fast-index",
"HypothesisNotes": "Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
"DegradationName": "warp-light-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 5,
"CorrectAnchors": 3,
"Precision": 0.6,
"Recall": 1,
"Confidence": 0.8353139386467288,
"TimingMeanAbsoluteErrorSamples": 44803.2,
"DecodeMilliseconds": 113.1098,
"RealtimeFactor": 4.420483459434991,
"IdentityScore": 0.8035313938646729,
"TimingScore": 0.005328218243819268,
"SpeedScore": 0.08840966918869983,
"TotalScore": 0.5168304773177246,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0003236487418848957,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 32,
"CepstralCoefficients": 10,
"TableCount": 3,
"HashBits": 12,
"NearHashRadius": 3,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 6,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "compact-fast-index/warp-light-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "compact-fast-index/warp-light-blur/post-warp.wav",
"ContentHash": "a7747c66b5bd084995a439c886b9867bf2578532f204d625e03772f58a31b16e"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "compact-fast-index/warp-light-blur/reconstructed-from-detections.wav",
"ContentHash": "fd56ad4755c04946b9cb4124f248bb31677d447c56f885a0610077130509ed00"
},
{
"Kind": "summary-json",
"Uri": "compact-fast-index/warp-light-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 0,
"Confidence": 0.95721979718768
},
{
"EventIndex": 13,
"SampleOffset": 9600,
"Confidence": 0.7213019744616194
},
{
"EventIndex": 17,
"SampleOffset": 13440,
"Confidence": 0.6408132491854348
},
{
"EventIndex": 1,
"SampleOffset": 15360,
"Confidence": 0.8834450524488017
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9737896199501074
}
]
}

View file

@ -0,0 +1,108 @@
{
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:warp-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "compact-fast-index",
"HypothesisNotes": "Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
"DegradationName": "warp-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 8,
"CorrectAnchors": 3,
"Precision": 0.375,
"Recall": 1,
"Confidence": 0.6273929348863668,
"TimingMeanAbsoluteErrorSamples": 278886,
"DecodeMilliseconds": 156.409,
"RealtimeFactor": 3.196746990262709,
"IdentityScore": 0.6814892934886367,
"TimingScore": 0.0008598267449109004,
"SpeedScore": 0.06393493980525418,
"TotalScore": 0.4354651187380895,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 9.12669517854302E-07,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 32,
"CepstralCoefficients": 10,
"TableCount": 3,
"HashBits": 12,
"NearHashRadius": 3,
"DenseStepSeconds": 0.04,
"ProposalBudgetMultiplier": 6,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "compact-fast-index/warp-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "compact-fast-index/warp-light/post-warp.wav",
"ContentHash": "fc695aa2e0579eab784d91523d9162deb58e036333ee6f19c185bdb90a449630"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "compact-fast-index/warp-light/reconstructed-from-detections.wav",
"ContentHash": "a5fc996352613ff2cf961cad93083a4e719cb918715d0d2546ffdf22c6d06923"
},
{
"Kind": "summary-json",
"Uri": "compact-fast-index/warp-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1920,
"Confidence": 0.9303400286319722
},
{
"EventIndex": 2,
"SampleOffset": 3840,
"Confidence": 0.8937756033130789
},
{
"EventIndex": 47,
"SampleOffset": 9600,
"Confidence": 0.4444455228149963
},
{
"EventIndex": 63,
"SampleOffset": 11520,
"Confidence": 0.36874646666158994
},
{
"EventIndex": 105,
"SampleOffset": 13440,
"Confidence": 0.2692392456861264
},
{
"EventIndex": 75,
"SampleOffset": 17280,
"Confidence": 0.34414717610057177
},
{
"EventIndex": 4,
"SampleOffset": 17904,
"Confidence": 0.8730382640410674
},
{
"EventIndex": 1,
"SampleOffset": 17952,
"Confidence": 0.8954111718415317
}
]
}

View file

@ -0,0 +1,84 @@
{
"ResultId": "bioacoustic-20260524-103438:highband-room-index:blur-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "highband-room-index",
"HypothesisNotes": "Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
"DegradationName": "blur-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9777684902055078,
"TimingMeanAbsoluteErrorSamples": 672,
"DecodeMilliseconds": 127.2228,
"RealtimeFactor": 3.930113155817982,
"IdentityScore": 0.9977768490205507,
"TimingScore": 0.2631578947368421,
"SpeedScore": 0.07860226311635964,
"TotalScore": 0.681710520068645,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0007356480619029866,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.035,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "highband-room-index/blur-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "highband-room-index/blur-light/post-warp.wav",
"ContentHash": "46984f1c8f4b37f309173dc8926769b24a95403f15b279da9d9fb1e6a1f971be"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "highband-room-index/blur-light/reconstructed-from-detections.wav",
"ContentHash": "515dd1055208673fdadb6db370480b779d056fdcc1ca43e6ccab2f577d8f0251"
},
{
"Kind": "summary-json",
"Uri": "highband-room-index/blur-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 3360,
"Confidence": 0.98044860544398
},
{
"EventIndex": 1,
"SampleOffset": 11760,
"Confidence": 0.9785031453289712
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9743537198435719
}
]
}

View file

@ -0,0 +1,89 @@
{
"ResultId": "bioacoustic-20260524-103438:highband-room-index:clean-roundtrip",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "highband-room-index",
"HypothesisNotes": "Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
"DegradationName": "clean-roundtrip",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.9482728927761414,
"TimingMeanAbsoluteErrorSamples": 5963.999999999999,
"DecodeMilliseconds": 123.188,
"RealtimeFactor": 4.058836899698022,
"IdentityScore": 0.8823272892776142,
"TimingScore": 0.03868471953578337,
"SpeedScore": 0.08117673799396044,
"TotalScore": 0.5702415164673539,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.008886063472396654,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.035,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "highband-room-index/clean-roundtrip/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "highband-room-index/clean-roundtrip/post-warp.wav",
"ContentHash": "f803fceaf796518f32920b9c777b27e7017379fcdbc6acb46e2c89889ef1aa74"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "highband-room-index/clean-roundtrip/reconstructed-from-detections.wav",
"ContentHash": "eadc71a086a97c50063fb45ca7df295f33a9e7634bb656ab7574c006c70cd6c4"
},
{
"Kind": "summary-json",
"Uri": "highband-room-index/clean-roundtrip/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 3360,
"Confidence": 0.9687346267460094
},
{
"EventIndex": 3,
"SampleOffset": 5040,
"Confidence": 0.8772393131190942
},
{
"EventIndex": 1,
"SampleOffset": 11760,
"Confidence": 0.976267321923383
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9708503093160791
}
]
}

View file

@ -0,0 +1,94 @@
{
"ResultId": "bioacoustic-20260524-103438:highband-room-index:warp-heavy-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "highband-room-index",
"HypothesisNotes": "Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
"DegradationName": "warp-heavy-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 5,
"CorrectAnchors": 3,
"Precision": 0.6,
"Recall": 1,
"Confidence": 0.8483369395650066,
"TimingMeanAbsoluteErrorSamples": 51724.8,
"DecodeMilliseconds": 154.5372,
"RealtimeFactor": 3.2354669296454186,
"IdentityScore": 0.8048336939565006,
"TimingScore": 0.0046185109920561605,
"SpeedScore": 0.06470933859290837,
"TotalScore": 0.5127700899501821,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.00033629216572444076,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.035,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "highband-room-index/warp-heavy-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "highband-room-index/warp-heavy-blur/post-warp.wav",
"ContentHash": "f9a185f36c36e210bd3673e625234a7ced9bfb7b199b08d83e11d4ec8ccd93e9"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "highband-room-index/warp-heavy-blur/reconstructed-from-detections.wav",
"ContentHash": "e7603a811dbdb2c3d2020462531857711bcb7d70bbefdd7686d778a22b8d0e69"
},
{
"Kind": "summary-json",
"Uri": "highband-room-index/warp-heavy-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1680,
"Confidence": 0.9533519594411115
},
{
"EventIndex": 31,
"SampleOffset": 6720,
"Confidence": 0.51949282981305
},
{
"EventIndex": 1,
"SampleOffset": 15120,
"Confidence": 0.9038583560223119
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9624784989041848
},
{
"EventIndex": 4,
"SampleOffset": 18192,
"Confidence": 0.9025030536443741
}
]
}

View file

@ -0,0 +1,84 @@
{
"ResultId": "bioacoustic-20260524-103438:highband-room-index:warp-light-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "highband-room-index",
"HypothesisNotes": "Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
"DegradationName": "warp-light-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9538086331550232,
"TimingMeanAbsoluteErrorSamples": 2352,
"DecodeMilliseconds": 141.2089,
"RealtimeFactor": 3.5408533031558207,
"IdentityScore": 0.9953808633155024,
"TimingScore": 0.09259259259259259,
"SpeedScore": 0.07081706606311641,
"TotalScore": 0.6479662151349015,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0003236487418848957,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.035,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "highband-room-index/warp-light-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "highband-room-index/warp-light-blur/post-warp.wav",
"ContentHash": "a7747c66b5bd084995a439c886b9867bf2578532f204d625e03772f58a31b16e"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "highband-room-index/warp-light-blur/reconstructed-from-detections.wav",
"ContentHash": "16197d44bb77722782f076946d787dcec030ad4aed5965dd5bef36aac6993f35"
},
{
"Kind": "summary-json",
"Uri": "highband-room-index/warp-light-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1680,
"Confidence": 0.9628247206431215
},
{
"EventIndex": 1,
"SampleOffset": 15120,
"Confidence": 0.9335501819466984
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9650509968752496
}
]
}

View file

@ -0,0 +1,94 @@
{
"ResultId": "bioacoustic-20260524-103438:highband-room-index:warp-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "highband-room-index",
"HypothesisNotes": "Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
"DegradationName": "warp-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 5,
"CorrectAnchors": 3,
"Precision": 0.6,
"Recall": 1,
"Confidence": 0.7705973915585707,
"TimingMeanAbsoluteErrorSamples": 95769.6,
"DecodeMilliseconds": 181.1372,
"RealtimeFactor": 2.760338572087898,
"IdentityScore": 0.797059739155857,
"TimingScore": 0.0024997500249975,
"SpeedScore": 0.05520677144175796,
"TotalScore": 0.5056683475694825,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 9.12669517854302E-07,
"Decoder": {
"FftSize": 1024,
"HopSize": 256,
"MelBins": 40,
"CepstralCoefficients": 14,
"TableCount": 4,
"HashBits": 14,
"NearHashRadius": 4,
"DenseStepSeconds": 0.035,
"ProposalBudgetMultiplier": 8,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "highband-room-index/warp-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "highband-room-index/warp-light/post-warp.wav",
"ContentHash": "fc695aa2e0579eab784d91523d9162deb58e036333ee6f19c185bdb90a449630"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "highband-room-index/warp-light/reconstructed-from-detections.wav",
"ContentHash": "4abc2f5c04b8d18b33f9292a6d6e08941c0b3ec248dea7fb23ebb4537654cf5c"
},
{
"Kind": "summary-json",
"Uri": "highband-room-index/warp-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 0,
"Confidence": 0.9243919896863865
},
{
"EventIndex": 1,
"SampleOffset": 8400,
"Confidence": 0.9089748749111662
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9277697421749646
},
{
"EventIndex": 30,
"SampleOffset": 18096,
"Confidence": 0.5637002965326965
},
{
"EventIndex": 35,
"SampleOffset": 18192,
"Confidence": 0.5281500544876397
}
]
}

View file

@ -0,0 +1,90 @@
{
"ResultId": "bioacoustic-20260524-103438:robust-wide-index:blur-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "robust-wide-index",
"HypothesisNotes": "More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
"DegradationName": "blur-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.9655822854926099,
"TimingMeanAbsoluteErrorSamples": 5016,
"DecodeMilliseconds": 220.4362,
"RealtimeFactor": 2.268229991262778,
"IdentityScore": 0.8840582285492611,
"TimingScore": 0.04566210045662101,
"SpeedScore": 0.04536459982525556,
"TotalScore": 0.5654081997477847,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0007356480619029866,
"Decoder": {
"FftSize": 1024,
"HopSize": 192,
"MelBins": 48,
"CepstralCoefficients": 18,
"TableCount": 5,
"HashBits": 15,
"NearHashRadius": 4,
"DenseStepSeconds": 0.03,
"ProposalBudgetMultiplier": 9,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur",
"template-warp-heavy"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "robust-wide-index/blur-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "robust-wide-index/blur-light/post-warp.wav",
"ContentHash": "46984f1c8f4b37f309173dc8926769b24a95403f15b279da9d9fb1e6a1f971be"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "robust-wide-index/blur-light/reconstructed-from-detections.wav",
"ContentHash": "5b9293a84d99d9d83cb4914b6a031365b64719af40f8f1eb372e8cf531091966"
},
{
"Kind": "summary-json",
"Uri": "robust-wide-index/blur-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 2880,
"Confidence": 0.984991303643638
},
{
"EventIndex": 1,
"SampleOffset": 10080,
"Confidence": 0.9775920300109124
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9830328920804959
},
{
"EventIndex": 4,
"SampleOffset": 18192,
"Confidence": 0.9167129162353933
}
]
}

View file

@ -0,0 +1,85 @@
{
"ResultId": "bioacoustic-20260524-103438:robust-wide-index:clean-roundtrip",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "robust-wide-index",
"HypothesisNotes": "More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
"DegradationName": "clean-roundtrip",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 3,
"CorrectAnchors": 3,
"Precision": 1,
"Recall": 1,
"Confidence": 0.9696885094694774,
"TimingMeanAbsoluteErrorSamples": 592,
"DecodeMilliseconds": 188.5116,
"RealtimeFactor": 2.6523566719501614,
"IdentityScore": 0.9969688509469478,
"TimingScore": 0.28846153846153844,
"SpeedScore": 0.053047133439003226,
"TotalScore": 0.6806531911979852,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.008886063472396654,
"Decoder": {
"FftSize": 1024,
"HopSize": 192,
"MelBins": 48,
"CepstralCoefficients": 18,
"TableCount": 5,
"HashBits": 15,
"NearHashRadius": 4,
"DenseStepSeconds": 0.03,
"ProposalBudgetMultiplier": 9,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur",
"template-warp-heavy"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "robust-wide-index/clean-roundtrip/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "robust-wide-index/clean-roundtrip/post-warp.wav",
"ContentHash": "f803fceaf796518f32920b9c777b27e7017379fcdbc6acb46e2c89889ef1aa74"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "robust-wide-index/clean-roundtrip/reconstructed-from-detections.wav",
"ContentHash": "6d83bb12aa7e63fa8e172ff14a5fbe0eb31141655097d8da27844bc167d4ecc9"
},
{
"Kind": "summary-json",
"Uri": "robust-wide-index/clean-roundtrip/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 4320,
"Confidence": 0.967394265955218
},
{
"EventIndex": 1,
"SampleOffset": 11520,
"Confidence": 0.9729967476603155
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9686745147928985
}
]
}

View file

@ -0,0 +1,90 @@
{
"ResultId": "bioacoustic-20260524-103438:robust-wide-index:warp-heavy-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "robust-wide-index",
"HypothesisNotes": "More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
"DegradationName": "warp-heavy-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.8346996513342604,
"TimingMeanAbsoluteErrorSamples": 57324,
"DecodeMilliseconds": 195.931,
"RealtimeFactor": 2.551918787736499,
"IdentityScore": 0.8709699651334262,
"TimingScore": 0.004169272461955389,
"SpeedScore": 0.05103837575472998,
"TotalScore": 0.5509595225768222,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.00033629216572444076,
"Decoder": {
"FftSize": 1024,
"HopSize": 192,
"MelBins": 48,
"CepstralCoefficients": 18,
"TableCount": 5,
"HashBits": 15,
"NearHashRadius": 4,
"DenseStepSeconds": 0.03,
"ProposalBudgetMultiplier": 9,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur",
"template-warp-heavy"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "robust-wide-index/warp-heavy-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "robust-wide-index/warp-heavy-blur/post-warp.wav",
"ContentHash": "f9a185f36c36e210bd3673e625234a7ced9bfb7b199b08d83e11d4ec8ccd93e9"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "robust-wide-index/warp-heavy-blur/reconstructed-from-detections.wav",
"ContentHash": "f0915094b4fe45e9f85b937593ee20b1641e65662acaf48e46129350a1214607"
},
{
"Kind": "summary-json",
"Uri": "robust-wide-index/warp-heavy-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 0,
"Confidence": 0.9528447339933648
},
{
"EventIndex": 1,
"SampleOffset": 10080,
"Confidence": 0.8853130095408102
},
{
"EventIndex": 30,
"SampleOffset": 11520,
"Confidence": 0.5397051505678829
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.960935711234984
}
]
}

View file

@ -0,0 +1,90 @@
{
"ResultId": "bioacoustic-20260524-103438:robust-wide-index:warp-light-blur",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "robust-wide-index",
"HypothesisNotes": "More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
"DegradationName": "warp-light-blur",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 4,
"CorrectAnchors": 3,
"Precision": 0.75,
"Recall": 1,
"Confidence": 0.9043167327507667,
"TimingMeanAbsoluteErrorSamples": 17004,
"DecodeMilliseconds": 202.6564,
"RealtimeFactor": 2.467230247848082,
"IdentityScore": 0.8779316732750768,
"TimingScore": 0.013917884481558803,
"SpeedScore": 0.049344604956961635,
"TotalScore": 0.5566917776286205,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 0.0003236487418848957,
"Decoder": {
"FftSize": 1024,
"HopSize": 192,
"MelBins": 48,
"CepstralCoefficients": 18,
"TableCount": 5,
"HashBits": 15,
"NearHashRadius": 4,
"DenseStepSeconds": 0.03,
"ProposalBudgetMultiplier": 9,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur",
"template-warp-heavy"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "robust-wide-index/warp-light-blur/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "robust-wide-index/warp-light-blur/post-warp.wav",
"ContentHash": "a7747c66b5bd084995a439c886b9867bf2578532f204d625e03772f58a31b16e"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "robust-wide-index/warp-light-blur/reconstructed-from-detections.wav",
"ContentHash": "0e7bc053eb411d43e7c134104a03f71f55d67e60b2092913e4866e990f0f7657"
},
{
"Kind": "summary-json",
"Uri": "robust-wide-index/warp-light-blur/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 1440,
"Confidence": 0.9563895832250255
},
{
"EventIndex": 1,
"SampleOffset": 8640,
"Confidence": 0.9299615644284689
},
{
"EventIndex": 9,
"SampleOffset": 11520,
"Confidence": 0.7591127388052877
},
{
"EventIndex": 2,
"SampleOffset": 17904,
"Confidence": 0.9718030445442848
}
]
}

View file

@ -0,0 +1,100 @@
{
"ResultId": "bioacoustic-20260524-103438:robust-wide-index:warp-light",
"RunId": "bioacoustic-20260524-103438",
"StartedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"Toolchain": "Mimir.BufferSmoke --bioacoustic-train",
"HypothesisId": "robust-wide-index",
"HypothesisNotes": "More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
"DegradationName": "warp-light",
"SampleRate": 48000,
"Seconds": 0.5,
"ExpectedEvents": 3,
"ObservationCount": 6,
"CorrectAnchors": 3,
"Precision": 0.5,
"Recall": 1,
"Confidence": 0.648330380997643,
"TimingMeanAbsoluteErrorSamples": 279728,
"DecodeMilliseconds": 231.5414,
"RealtimeFactor": 2.159441033007488,
"IdentityScore": 0.7398330380997643,
"TimingScore": 0.0008572408275231455,
"SpeedScore": 0.04318882066014976,
"TotalScore": 0.467488551102838,
"RoundTripMelBins": 48,
"RoundTripCepstralCoefficients": 20,
"RoundTripFrameCount": 43,
"RoundTripRmsRatio": 9.12669517854302E-07,
"Decoder": {
"FftSize": 1024,
"HopSize": 192,
"MelBins": 48,
"CepstralCoefficients": 18,
"TableCount": 5,
"HashBits": 15,
"NearHashRadius": 4,
"DenseStepSeconds": 0.03,
"ProposalBudgetMultiplier": 9,
"TemplateAugmentations": [
"template-clean",
"template-blur",
"template-warp-light",
"template-warp-blur",
"template-warp-heavy"
]
},
"Artifacts": [
{
"Kind": "pre-warp-audio",
"Uri": "robust-wide-index/warp-light/pre-warp.wav",
"ContentHash": "e72bae7434ebccd1826ae87389724ae617b5158316e80fd90a35d4d7a90c87f2"
},
{
"Kind": "post-warp-audio",
"Uri": "robust-wide-index/warp-light/post-warp.wav",
"ContentHash": "fc695aa2e0579eab784d91523d9162deb58e036333ee6f19c185bdb90a449630"
},
{
"Kind": "reconstructed-from-detections",
"Uri": "robust-wide-index/warp-light/reconstructed-from-detections.wav",
"ContentHash": "487542b43094089b182bb4ed206d9e150dfd254af33c59ca7af574037908a787"
},
{
"Kind": "summary-json",
"Uri": "robust-wide-index/warp-light/summary.json",
"ContentHash": ""
}
],
"Observations": [
{
"EventIndex": 0,
"SampleOffset": 2880,
"Confidence": 0.9331180942916268
},
{
"EventIndex": 1,
"SampleOffset": 8640,
"Confidence": 0.9169619441254396
},
{
"EventIndex": 73,
"SampleOffset": 12960,
"Confidence": 0.34169144863143375
},
{
"EventIndex": 30,
"SampleOffset": 14400,
"Confidence": 0.5427242045698007
},
{
"EventIndex": 119,
"SampleOffset": 15840,
"Confidence": 0.24752983653104574
},
{
"EventIndex": 2,
"SampleOffset": 17952,
"Confidence": 0.9079567578365106
}
]
}

View file

@ -0,0 +1,19 @@
{
"runId": "bioacoustic-20260524-103438",
"startedAtUtc": "2026-05-24T10:34:38.4598729\u002B00:00",
"sampleRate": 48000,
"seconds": 0.5,
"resultCount": 20,
"cache": "bioacoustic-training.cc",
"best": {
"ResultId": "bioacoustic-20260524-103438:compact-fast-index:clean-roundtrip",
"HypothesisId": "compact-fast-index",
"DegradationName": "clean-roundtrip",
"TotalScore": 0.6995775612751758,
"IdentityScore": 0.9973468589769867,
"TimingScore": 0.35714285714285715,
"SpeedScore": 0.08468397211864902,
"DecodeMilliseconds": 118.0861,
"RealtimeFactor": 4.234198605932451
}
}

View file

@ -99,6 +99,10 @@ Expected state:
- `--bioacoustic-cepstral-smoke` round-trips the active call through a degraded
mel-cepstral representation, then decodes word identity with an augmented
projection-hash MFCC index instead of brute-forcing every word waveform.
- `--bioacoustic-train` runs a hypothesis panel over decoder shapes and writes
CultCache receipts plus pre-warp, post-warp, and reconstructed-from-detections
WAV artifacts. See [[bioacoustic-training-harness|Bioacoustic Training
Harness]].
Render a raw Float32 preview:
@ -126,3 +130,8 @@ audio window
Do not add a pile of threshold rules to make one room capture look good. The
right extension is learned path weighting over motif parts: which contours,
partials, and time gaps survive each output/mic path.
The latest training receipt says identity can survive the current
mel-cepstral degradation panel, but timing collapses under warped domains. That
points the next cut at a global delay/clock/path hypothesis over detected words,
not a larger brute-force dictionary.

View file

@ -0,0 +1,84 @@
# Bioacoustic Training Harness
## Purpose
The bioacoustic receiver is tuned by benchmark receipts, not by console optimism.
`Mimir.BufferSmoke --bioacoustic-train` runs a small hypothesis panel over the
current motif language and records both queryable CultCache evidence and audio
artifacts.
This is authoring and calibration machinery. It is allowed to allocate, score,
serialize, and leave receipts. It is not the runtime hot loop.
## Command
```powershell
dotnet run --project .\src\Mimir.BufferSmoke\Mimir.BufferSmoke.csproj -- --bioacoustic-train --sample-rate 48000 --seconds 0.5 --output artifacts/bioacoustic-training
```
Each run writes:
- `bioacoustic-training.cc`: typed CultCache results with provenance, decoder
settings, degradation name, scores, timing, and artifact hashes.
- `run-summary.json`: short best-result summary.
- `source-pre-warp.wav`: the emitted source song.
- `<hypothesis>/<degradation>/pre-warp.wav`: source audio for that scored case.
- `<hypothesis>/<degradation>/post-warp.wav`: mel-cepstral warp/blur round-trip.
- `<hypothesis>/<degradation>/reconstructed-from-detections.wav`: song rebuilt
from detected word features.
- `<hypothesis>/<degradation>/summary.json`: per-case receipt mirror.
## Current Hypotheses
- `baseline-mfcc-index`: current indexed MFCC/log-mel smoke shape.
- `compact-fast-index`: smaller feature and hash surface to test speed against
identity loss.
- `robust-wide-index`: wider cepstral/augmentation surface to test degraded-path
survival.
- `highband-room-index`: upper-band-biased receiver to avoid low room junk.
The harness intentionally scores identity, timing, and speed separately:
```text
total = identity * 0.62 + timing * 0.18 + speed * 0.20
```
Aggregate score is a triage knife. Identity tells whether the word language
survived. Timing tells whether detected words landed on the canonical clock.
Speed tells whether the receiver shape can plausibly become a streaming machine.
## Latest Local Receipt
Run:
```text
artifacts/bioacoustic-training/bioacoustic-20260524-103438/
```
Best aggregate result:
```text
compact-fast-index / clean-roundtrip
total=0.700 identity=0.997 timing=0.357 speed=0.085 realtime=4.2x
```
Useful observations:
- `compact-fast-index` is faster on clean/light blur, but collapses under warp.
- `baseline-mfcc-index` keeps better identity under light warp than compact.
- `highband-room-index` wins some blur/warp-blur cases, which supports
room-path band selection as a real tuning axis.
- `robust-wide-index` is not currently worth its extra CPU except as pressure
for heavier blur/warp cases.
- Timing mean absolute error is still poor under warped cepstral reconstruction.
That means the next cut is global delay/clock/path hypothesis over detected
words, not just “more cepstra.”
## Architecture Pressure
The benchmark now uses CultLib at the BufferSmoke edge, but Mimir itself should
be a CultMesh app. Phones, microcontrollers, Raven, and Starfire need typed
state surfaces for schedule/codebook snapshots, calibration receipts, learned
path response, and work offload. CultCache receipts are the durable proof layer;
the next runtime cut should promote the live state contracts into mesh-owned
documents rather than leaving them as harness-only artifacts.

View file

@ -48,7 +48,9 @@ a named invariant that the native runtime cannot protect yet.
to render the active motif timeline into memory and verify that the decoder
recovers direct word anchors. Use `--standalone-bioacoustic-self-test` to
verify that a receiver with only the codebook/schedule can recover canonical
source offset from delayed audio.
source offset from delayed audio. Use `--bioacoustic-train` to run the
indexed cepstral receiver hypothesis panel and write CultCache/audio receipts
under `artifacts/bioacoustic-training/`.
- `native/probes/wasapi_audio_cadence` captures WASAPI mic or render-loopback
block metadata and emits `audio-block` JSON events for the diagnostic runtime
adapter. It can probe requested shared/exclusive formats so driver state is
@ -187,7 +189,9 @@ a named invariant that the native runtime cannot protect yet.
contour, formant, band, gain, rhythm weighting, and left/right speaker
separation without weakening the standalone word-identity receiver invariant.
Keep chirp-bin calibration artifacts as reference data, not the runtime
target.
target. Promote the `--bioacoustic-train` receipt shape into a real
calibration loop: detected words should feed a global delay/clock/path
hypothesis and learned per-output/mic response, not per-case threshold rules.
4. Add the synchronization actuator: drive a variable-rate resampler and
fractional delay line per non-reference stream from the smoothed
`MimirAudioSynchronizationState`. First, prove the bioacoustic motif decoder

View file

@ -35,6 +35,10 @@ Ownership:
- Mimir owns configuration, calibration truth, launch, status, persistence, and
runtime contracts.
- Mimir is a CultMesh app: typed state surfaces for codebooks, schedules,
calibration receipts, path response, and offload work must be mesh-syncable so
phones, microcontrollers, Raven, and Starfire can participate without
becoming independent clock authorities.
- `Mimir.Runtime` owns app-level stream buffers and synchronization.
- Native capture workers own device reads.
- Fensalir owns dense visual fusion, material/brush/splat reconciliation,
@ -131,6 +135,13 @@ proves a receiver with only codebook/schedule state can recover delayed source
time to below printed microsecond precision. `--chirp-only-sync-self-test
--sample-rate 48000` recovers a 317.375-sample synthetic delay with printed
0.000 us error using `evidence=bioacoustic`.
`--bioacoustic-train` is now the receipt-backed tuning harness: it runs
multiple indexed cepstral decoder hypotheses across mel-cepstral warp/blur
degradations and writes typed CultCache results plus pre-warp, post-warp, and
reconstructed-from-detections WAV artifacts. The latest local receipt under
`artifacts/bioacoustic-training/bioacoustic-20260524-103438/` shows identity
survives many degradations, but timing still fails under warped domains, so the
next receiver cut is a global delay/clock/path hypothesis over detected words.
Reports now carry fractional delay and per-band matched energy. The older
`MimirChirpletSymbolCodebook` / `MimirChirpletStreamDecoder` path remains a
diagnostic reference for constrained chirplet-transform work. The active

View file

@ -10,6 +10,7 @@
</PropertyGroup>
<ItemGroup>
<ProjectReference Include="..\..\..\CultLib\src\GameCult.Caching.MessagePack\GameCult.Caching.MessagePack.csproj" />
<ProjectReference Include="..\Mimir.Runtime\Mimir.Runtime.csproj" />
</ItemGroup>

View file

@ -1,5 +1,10 @@
using System.Diagnostics;
using System.Numerics;
using System.Security.Cryptography;
using System.Text.Json;
using GameCult.Caching;
using GameCult.Caching.MessagePack;
using MessagePack;
using Mimir.Runtime.Synchronization;
if (args.Any(arg => string.Equals(arg, "--chirplet-self-test", StringComparison.OrdinalIgnoreCase)))
@ -31,6 +36,14 @@ if (args.Any(arg => string.Equals(arg, "--bioacoustic-cepstral-smoke", StringCom
ParseDoubleOption(args, "--seconds", 0.75));
}
if (args.Any(arg => string.Equals(arg, "--bioacoustic-train", StringComparison.OrdinalIgnoreCase)))
{
return await RunBioacousticTrainingAsync(
ParseIntOption(args, "--sample-rate", MimirBioacousticTimeline.SampleRate),
ParseDoubleOption(args, "--seconds", 0.75),
ParseStringOption(args, "--output", "artifacts/bioacoustic-training")).ConfigureAwait(false);
}
if (args.Any(arg => string.Equals(arg, "--standalone-chirp-bin-self-test", StringComparison.OrdinalIgnoreCase)))
{
return RunStandaloneChirpBinSelfTest(
@ -480,6 +493,180 @@ static int RunBioacousticCepstralSmoke(int sampleRate, double seconds)
return 0;
}
static async Task<int> RunBioacousticTrainingAsync(int sampleRate, double seconds, string outputRoot)
{
var runStarted = DateTimeOffset.UtcNow;
var runId = $"bioacoustic-{runStarted:yyyyMMdd-HHmmss}";
var runDirectory = Path.GetFullPath(Path.Combine(outputRoot, runId));
Directory.CreateDirectory(runDirectory);
var timeline = MimirBioacousticTimeline.Default;
var segmentCount = Math.Max(1, (int)Math.Ceiling(seconds / MimirBioacousticTimeline.SegmentSeconds));
var samples = new List<float>(Math.Max(1, (int)Math.Ceiling(seconds * sampleRate)));
for (ulong segment = 0; segment < (ulong)segmentCount; segment++)
{
samples.AddRange(timeline.RenderSegmentMonoFloat(segment, sampleRate));
}
var requestedSamples = Math.Max(1, (int)Math.Round(seconds * sampleRate));
if (samples.Count > requestedSamples)
{
samples.RemoveRange(requestedSamples, samples.Count - requestedSamples);
}
var source = samples.ToArray();
var degradations = new[]
{
new CepstralDegradationSetting("clean-roundtrip", 0.0, 0.0, 0),
new CepstralDegradationSetting("blur-light", 0.0, 0.0, 1),
new CepstralDegradationSetting("warp-light", 0.75, 1.25, 0),
new CepstralDegradationSetting("warp-light-blur", 0.75, 1.25, 1),
new CepstralDegradationSetting("warp-heavy-blur", 1.25, 1.75, 1),
};
var hypotheses = BioacousticTrainingHypotheses();
var expectedEvents = timeline.EventsOverlapping(0.0, source.Length / (double)sampleRate)
.Select(timelineEvent => timelineEvent.Index)
.ToHashSet();
var cachePath = Path.Combine(runDirectory, "bioacoustic-training.cc");
var results = new List<BioacousticTrainingResult>();
WriteWave(Path.Combine(runDirectory, "source-pre-warp.wav"), source, sampleRate);
foreach (var hypothesis in hypotheses)
{
var hypothesisDirectory = Path.Combine(runDirectory, hypothesis.Id);
Directory.CreateDirectory(hypothesisDirectory);
var templateIndex = BuildCepstralWordIndex(sampleRate, hypothesis.Decoder);
foreach (var degradation in degradations)
{
var settingDirectory = Path.Combine(hypothesisDirectory, degradation.Name);
Directory.CreateDirectory(settingDirectory);
var preWarpPath = Path.Combine(settingDirectory, "pre-warp.wav");
var postWarpPath = Path.Combine(settingDirectory, "post-warp.wav");
var reconstructedPath = Path.Combine(settingDirectory, "reconstructed-from-detections.wav");
var summaryPath = Path.Combine(settingDirectory, "summary.json");
WriteWave(preWarpPath, source, sampleRate);
var degraded = RoundTripThroughDegradedCepstrum(source, sampleRate, degradation, out var analysis);
WriteWave(postWarpPath, degraded, sampleRate);
var decodeStamp = Stopwatch.GetTimestamp();
var observations = DecodeCepstralIndexedWordsWithIndex(degraded, sampleRate, expectedEvents.Count, hypothesis.Decoder, templateIndex);
var decodeMilliseconds = Stopwatch.GetElapsedTime(decodeStamp).TotalMilliseconds;
var correctAnchors = observations.Count(observation => expectedEvents.Contains(observation.EventIndex));
var precision = observations.Count == 0 ? 0.0 : correctAnchors / (double)observations.Count;
var recall = expectedEvents.Count == 0 ? 0.0 : correctAnchors / (double)expectedEvents.Count;
var confidence = observations.Count == 0 ? 0.0 : observations.Average(observation => observation.Confidence);
var timingMae = MeanCepstralTimingError(observations, timeline, sampleRate);
var realtime = decodeMilliseconds <= 0.0 ? double.PositiveInfinity : seconds * 1000.0 / decodeMilliseconds;
var identityScore = precision * 0.45 + recall * 0.45 + confidence * 0.10;
var timingScore = double.IsFinite(timingMae)
? 1.0 / (1.0 + timingMae / Math.Max(1.0, sampleRate * 0.005))
: 0.0;
var speedScore = Math.Clamp(realtime / 50.0, 0.0, 1.0);
var totalScore = identityScore * 0.62 + timingScore * 0.18 + speedScore * 0.20;
var reconstructed = ReconstructDetectedBioacousticSong(observations, source.Length, sampleRate);
WriteWave(reconstructedPath, reconstructed, sampleRate);
var artifactRoot = RelativePath(runDirectory, settingDirectory);
var result = new BioacousticTrainingResult(
$"{runId}:{hypothesis.Id}:{degradation.Name}",
runId,
runStarted.ToString("O"),
"Mimir.BufferSmoke --bioacoustic-train",
hypothesis.Id,
hypothesis.Notes,
degradation.Name,
sampleRate,
seconds,
expectedEvents.Count,
observations.Count,
correctAnchors,
precision,
recall,
confidence,
timingMae,
decodeMilliseconds,
realtime,
identityScore,
timingScore,
speedScore,
totalScore,
analysis.MelBins,
analysis.CepstralCoefficients,
analysis.FrameCount,
analysis.RmsRatio,
new BioacousticTrainingDecoderSnapshot(
hypothesis.Decoder.FftSize,
hypothesis.Decoder.HopSize,
hypothesis.Decoder.MelBins,
hypothesis.Decoder.CepstralCoefficients,
hypothesis.Decoder.TableCount,
hypothesis.Decoder.HashBits,
hypothesis.Decoder.NearHashRadius,
hypothesis.Decoder.DenseStepSeconds,
hypothesis.Decoder.ProposalBudgetMultiplier,
hypothesis.Decoder.TemplateAugmentations.Select(setting => setting.Name).ToArray()),
[
new BioacousticTrainingArtifact("pre-warp-audio", $"{artifactRoot}/pre-warp.wav", Sha256File(preWarpPath)),
new BioacousticTrainingArtifact("post-warp-audio", $"{artifactRoot}/post-warp.wav", Sha256File(postWarpPath)),
new BioacousticTrainingArtifact("reconstructed-from-detections", $"{artifactRoot}/reconstructed-from-detections.wav", Sha256File(reconstructedPath)),
new BioacousticTrainingArtifact("summary-json", $"{artifactRoot}/summary.json", "")
],
observations
.Select(observation => new BioacousticTrainingObservation(
observation.EventIndex,
observation.SampleOffset,
observation.Confidence))
.ToArray());
await File.WriteAllTextAsync(
summaryPath,
JsonSerializer.Serialize(result, new JsonSerializerOptions { WriteIndented = true })).ConfigureAwait(false);
result = result with
{
Artifacts = result.Artifacts
.Select(artifact => artifact.Kind == "summary-json"
? artifact with { ContentHash = Sha256File(summaryPath) }
: artifact)
.ToArray()
};
results.Add(result);
Console.WriteLine(
$"bioacoustic-train hypothesis={hypothesis.Id} degradation={degradation.Name} total={totalScore:0.000} identity={identityScore:0.000} timing={timingScore:0.000} speed={speedScore:0.000} " +
$"precision={precision:0.000} recall={recall:0.000} confidence={confidence:0.000} mae={timingMae:0.000} decodeMs={decodeMilliseconds:0.000} realtime={realtime:0.0}x artifacts={RelativePath(Directory.GetCurrentDirectory(), settingDirectory)}");
}
}
await WriteBioacousticTrainingCacheAsync(cachePath, results).ConfigureAwait(false);
var best = results.OrderByDescending(result => result.TotalScore).First();
var runSummaryPath = Path.Combine(runDirectory, "run-summary.json");
await File.WriteAllTextAsync(
runSummaryPath,
JsonSerializer.Serialize(new
{
runId,
startedAtUtc = runStarted.ToString("O"),
sampleRate,
seconds,
resultCount = results.Count,
cache = Path.GetFileName(cachePath),
best = new
{
best.ResultId,
best.HypothesisId,
best.DegradationName,
best.TotalScore,
best.IdentityScore,
best.TimingScore,
best.SpeedScore,
best.DecodeMilliseconds,
best.RealtimeFactor
}
}, new JsonSerializerOptions { WriteIndented = true })).ConfigureAwait(false);
Console.WriteLine($"bioacoustic-training-run path={runDirectory} cache={cachePath} results={results.Count} best={best.HypothesisId}/{best.DegradationName} total={best.TotalScore:0.000}");
return 0;
}
static int RunPassiveSyncSelfTest()
{
const int sampleRate = 48_000;
@ -702,6 +889,76 @@ static int RenderBioacousticFloat32(string outputPath, int sampleRate, double se
return 0;
}
static IReadOnlyList<BioacousticTrainingHypothesis> BioacousticTrainingHypotheses()
{
var clean = new CepstralDegradationSetting("template-clean", 0.0, 0.0, 0);
var blur = new CepstralDegradationSetting("template-blur", 0.0, 0.0, 1);
var warp = new CepstralDegradationSetting("template-warp-light", 0.75, 1.25, 0);
var warpBlur = new CepstralDegradationSetting("template-warp-blur", 0.75, 1.25, 1);
return
[
new(
"baseline-mfcc-index",
"Current smoke baseline: four projection tables, 14-bit hashes, light degradation augmentation.",
CepstralDecoderOptions.Default),
new(
"compact-fast-index",
"Smaller feature and table shape. It should run faster if identity survives; otherwise it dies with a receipt.",
new CepstralDecoderOptions(
FftSize: 1024,
HopSize: 256,
MelBins: 32,
CepstralCoefficients: 10,
MinFrequencyHz: 300,
MaxFrequencyHz: 14_500,
TableCount: 3,
HashBits: 12,
NearHashRadius: 3,
DenseStepSeconds: 0.040,
ProposalBudgetMultiplier: 6.0,
TemplateAugmentations: [clean, blur, warpBlur])),
new(
"robust-wide-index",
"More cepstra and a wider augmentation bank. It pays more CPU for warped-room survival.",
new CepstralDecoderOptions(
FftSize: 1024,
HopSize: 192,
MelBins: 48,
CepstralCoefficients: 18,
MinFrequencyHz: 180,
MaxFrequencyHz: 15_500,
TableCount: 5,
HashBits: 15,
NearHashRadius: 4,
DenseStepSeconds: 0.030,
ProposalBudgetMultiplier: 9.0,
TemplateAugmentations:
[
clean,
blur,
warp,
warpBlur,
new("template-warp-heavy", 1.25, 1.75, 1)
])),
new(
"highband-room-index",
"Bias toward the measured useful upper speech/birdcall band and away from low room junk.",
new CepstralDecoderOptions(
FftSize: 1024,
HopSize: 256,
MelBins: 40,
CepstralCoefficients: 14,
MinFrequencyHz: 1_200,
MaxFrequencyHz: 16_000,
TableCount: 4,
HashBits: 14,
NearHashRadius: 4,
DenseStepSeconds: 0.035,
ProposalBudgetMultiplier: 8.0,
TemplateAugmentations: [clean, blur, warp, warpBlur]))
];
}
static MimirChirpBinCalibrationModel? LoadOptionalCalibration(string[] args)
{
var path = ParseStringOption(args, "--calibration", "");
@ -1048,9 +1305,26 @@ static float[] RoundTripThroughDegradedCepstrum(
static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWords(float[] samples, int sampleRate, int expectedEventCount)
{
const int tableCount = 4;
const int hashBits = 14;
var templateIndex = BuildCepstralWordIndex(sampleRate, tableCount, hashBits);
return DecodeCepstralIndexedWordsWithOptions(samples, sampleRate, expectedEventCount, CepstralDecoderOptions.Default);
}
static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWordsWithOptions(
float[] samples,
int sampleRate,
int expectedEventCount,
CepstralDecoderOptions options)
{
var templateIndex = BuildCepstralWordIndex(sampleRate, options);
return DecodeCepstralIndexedWordsWithIndex(samples, sampleRate, expectedEventCount, options, templateIndex);
}
static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWordsWithIndex(
float[] samples,
int sampleRate,
int expectedEventCount,
CepstralDecoderOptions options,
CepstralWordIndex templateIndex)
{
var motifSamples = MimirBioacousticTimeline.Default.RenderEventMonoFloat(0, sampleRate).Length;
var hopSamples = Math.Max(1, sampleRate / 1_000);
var energyTrace = WindowEnergy(samples, motifSamples, hopSamples);
@ -1068,13 +1342,13 @@ static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWords(float[]
}
}
var denseStep = Math.Max(1, (int)Math.Round(sampleRate * 0.040));
var denseStep = Math.Max(1, (int)Math.Round(sampleRate * options.DenseStepSeconds));
for (var offset = 0; offset + motifSamples <= samples.Length; offset += denseStep)
{
proposals.Add(offset);
}
var proposalBudget = Math.Max(expectedEventCount * 8, 16);
var proposalBudget = Math.Max((int)Math.Ceiling(expectedEventCount * options.ProposalBudgetMultiplier), 16);
var observations = new List<CepstralWordObservation>();
foreach (var offset in proposals
.OrderByDescending(offset => energyTrace[Math.Clamp(offset / hopSamples, 0, energyTrace.Length - 1)])
@ -1086,11 +1360,11 @@ static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWords(float[]
continue;
}
var feature = CepstralFingerprint(samples.AsSpan(offset, motifSamples), sampleRate);
var feature = CepstralFingerprintWithOptions(samples.AsSpan(offset, motifSamples), sampleRate, options);
var candidateIndexes = new HashSet<int>();
for (var table = 0; table < tableCount; table++)
for (var table = 0; table < options.TableCount; table++)
{
var key = ProjectionHash(feature, table, hashBits);
var key = ProjectionHash(feature, table, options.HashBits);
if (templateIndex.Buckets.TryGetValue((table, key), out var bucket))
{
foreach (var candidate in bucket)
@ -1103,7 +1377,7 @@ static IReadOnlyList<CepstralWordObservation> DecodeCepstralIndexedWords(float[]
{
foreach (var near in templateIndex.Buckets
.Where(pair => pair.Key.Table == table &&
HammingDistance(pair.Key.Hash, key) <= 4)
HammingDistance(pair.Key.Hash, key) <= options.NearHashRadius)
.SelectMany(pair => pair.Value))
{
candidateIndexes.Add(near);
@ -1154,34 +1428,27 @@ static ulong PredictedBioacousticEventIndex(double sampleOffset, int sampleRate)
return (ulong)Math.Max(0, index);
}
static CepstralWordIndex BuildCepstralWordIndex(int sampleRate, int tableCount, int hashBits)
static CepstralWordIndex BuildCepstralWordIndex(int sampleRate, CepstralDecoderOptions options)
{
var augmentationSettings = new[]
{
new CepstralDegradationSetting("template-clean", 0.0, 0.0, 0),
new CepstralDegradationSetting("template-blur", 0.0, 0.0, 1),
new CepstralDegradationSetting("template-warp-light", 0.75, 1.25, 0),
new CepstralDegradationSetting("template-warp-blur", 0.75, 1.25, 1),
};
var templates = new List<CepstralWordTemplate>(MimirBioacousticTimeline.SymbolCount);
for (ulong eventIndex = 0; eventIndex < MimirBioacousticTimeline.SymbolCount; eventIndex++)
{
var samples = MimirBioacousticTimeline.Default.RenderEventMonoFloat(eventIndex, sampleRate);
foreach (var setting in augmentationSettings)
foreach (var setting in options.TemplateAugmentations)
{
var templateSamples = setting.BlurPasses == 0 && setting.WarpFrames == 0.0 && setting.WarpCoefficients == 0.0
? samples
: RoundTripThroughDegradedCepstrum(samples, sampleRate, setting, out _);
templates.Add(new CepstralWordTemplate(eventIndex, CepstralFingerprint(templateSamples, sampleRate)));
templates.Add(new CepstralWordTemplate(eventIndex, CepstralFingerprintWithOptions(templateSamples, sampleRate, options)));
}
}
var buckets = new Dictionary<(int Table, int Hash), List<int>>();
for (var index = 0; index < templates.Count; index++)
{
for (var table = 0; table < tableCount; table++)
for (var table = 0; table < options.TableCount; table++)
{
var key = (table, ProjectionHash(templates[index].Feature, table, hashBits));
var key = (table, ProjectionHash(templates[index].Feature, table, options.HashBits));
if (!buckets.TryGetValue(key, out var bucket))
{
bucket = [];
@ -1195,24 +1462,20 @@ static CepstralWordIndex BuildCepstralWordIndex(int sampleRate, int tableCount,
return new CepstralWordIndex(templates.ToArray(), buckets);
}
static double[] CepstralFingerprint(ReadOnlySpan<float> samples, int sampleRate)
static double[] CepstralFingerprintWithOptions(ReadOnlySpan<float> samples, int sampleRate, CepstralDecoderOptions options)
{
const int fftSize = 1024;
const int hopSize = 256;
const int melBins = 40;
const int cepstralCoefficients = 14;
var window = HannWindow(fftSize);
var melFilters = BuildMelFilterBank(melBins, fftSize, sampleRate, 180.0, 15_000.0);
var window = HannWindow(options.FftSize);
var melFilters = BuildMelFilterBank(options.MelBins, options.FftSize, sampleRate, options.MinFrequencyHz, options.MaxFrequencyHz);
var melNormalizer = melFilters.Select(filter => Math.Max(1.0e-12, filter.Sum())).ToArray();
var frameCount = Math.Max(1, 1 + Math.Max(0, samples.Length - fftSize) / hopSize);
var mean = new double[cepstralCoefficients];
var delta = new double[cepstralCoefficients];
var frameCount = Math.Max(1, 1 + Math.Max(0, samples.Length - options.FftSize) / options.HopSize);
var mean = new double[options.CepstralCoefficients];
var delta = new double[options.CepstralCoefficients];
double[]? previous = null;
for (var frame = 0; frame < frameCount; frame++)
{
var offset = frame * hopSize;
var spectrum = new Complex[fftSize];
for (var index = 0; index < fftSize; index++)
var offset = frame * options.HopSize;
var spectrum = new Complex[options.FftSize];
for (var index = 0; index < options.FftSize; index++)
{
var sampleIndex = offset + index;
var sample = sampleIndex < samples.Length ? samples[sampleIndex] : 0.0f;
@ -1220,8 +1483,8 @@ static double[] CepstralFingerprint(ReadOnlySpan<float> samples, int sampleRate)
}
FastFourierTransform(spectrum, inverse: false);
var cepstrum = Dct(SpectrumToLogMel(spectrum, melFilters, melNormalizer), cepstralCoefficients);
for (var coefficient = 0; coefficient < cepstralCoefficients; coefficient++)
var cepstrum = Dct(SpectrumToLogMel(spectrum, melFilters, melNormalizer), options.CepstralCoefficients);
for (var coefficient = 0; coefficient < options.CepstralCoefficients; coefficient++)
{
mean[coefficient] += cepstrum[coefficient];
if (previous != null)
@ -1233,11 +1496,11 @@ static double[] CepstralFingerprint(ReadOnlySpan<float> samples, int sampleRate)
previous = cepstrum;
}
var output = new double[cepstralCoefficients * 2];
for (var coefficient = 0; coefficient < cepstralCoefficients; coefficient++)
var output = new double[options.CepstralCoefficients * 2];
for (var coefficient = 0; coefficient < options.CepstralCoefficients; coefficient++)
{
output[coefficient] = coefficient == 0 ? 0.0 : mean[coefficient] / frameCount;
output[coefficient + cepstralCoefficients] = delta[coefficient] / Math.Max(1, frameCount - 1);
output[coefficient + options.CepstralCoefficients] = delta[coefficient] / Math.Max(1, frameCount - 1);
}
NormalizeFeature(output);
@ -1756,6 +2019,80 @@ static void AppendFloatBlock(MimirRollingStreamBuffer buffer, string sourceId, f
1_000_000_000L)));
}
static float[] ReconstructDetectedBioacousticSong(IReadOnlyList<CepstralWordObservation> observations, int sampleCount, int sampleRate)
{
var output = new float[sampleCount];
foreach (var observation in observations)
{
var eventSamples = MimirBioacousticTimeline.Default.RenderEventMonoFloat(observation.EventIndex, sampleRate);
var offset = (int)Math.Round(observation.SampleOffset);
for (var index = 0; index < eventSamples.Length; index++)
{
var target = offset + index;
if (target >= 0 && target < output.Length)
{
output[target] += eventSamples[index] * (float)Math.Clamp(observation.Confidence, 0.0, 1.0);
}
}
}
return output;
}
static async Task WriteBioacousticTrainingCacheAsync(string cachePath, IReadOnlyList<BioacousticTrainingResult> results)
{
using var cache = await CultCacheMessagePack.OpenAsync(cachePath, new CultCacheOpenOptions
{
PullOnOpen = File.Exists(cachePath)
}).ConfigureAwait(false);
foreach (var result in results)
{
await cache.UpsertAsync(
result,
new CultRecordHandle<BioacousticTrainingResult>(new CultRecordKey($"bioacoustic-training-result:{result.ResultId}")))
.ConfigureAwait(false);
}
await cache.FlushAsync().ConfigureAwait(false);
}
static void WriteWave(string path, IReadOnlyList<float> samples, int sampleRate)
{
Directory.CreateDirectory(Path.GetDirectoryName(Path.GetFullPath(path)) ?? ".");
using var stream = File.Create(path);
using var writer = new BinaryWriter(stream);
var dataBytes = samples.Count * sizeof(short);
writer.Write("RIFF"u8);
writer.Write(36 + dataBytes);
writer.Write("WAVE"u8);
writer.Write("fmt "u8);
writer.Write(16);
writer.Write((short)1);
writer.Write((short)1);
writer.Write(sampleRate);
writer.Write(sampleRate * sizeof(short));
writer.Write((short)sizeof(short));
writer.Write((short)16);
writer.Write("data"u8);
writer.Write(dataBytes);
for (var index = 0; index < samples.Count; index++)
{
writer.Write((short)Math.Clamp(Math.Round(samples[index] * short.MaxValue), short.MinValue, short.MaxValue));
}
}
static string Sha256File(string path)
{
using var stream = File.OpenRead(path);
return Convert.ToHexString(SHA256.HashData(stream)).ToLowerInvariant();
}
static string RelativePath(string root, string path)
{
var relative = Path.GetRelativePath(root, path).Replace('\\', '/');
return relative == "." ? "" : relative;
}
internal sealed class DisposableConfiguration : IDisposable
{
private readonly List<IMimirStreamSource> ownedSources;
@ -1788,6 +2125,46 @@ internal sealed record CepstralDegradationSetting(
double WarpCoefficients,
int BlurPasses);
internal sealed record CepstralDecoderOptions(
int FftSize,
int HopSize,
int MelBins,
int CepstralCoefficients,
double MinFrequencyHz,
double MaxFrequencyHz,
int TableCount,
int HashBits,
int NearHashRadius,
double DenseStepSeconds,
double ProposalBudgetMultiplier,
IReadOnlyList<CepstralDegradationSetting> TemplateAugmentations)
{
public static CepstralDecoderOptions Default { get; } = new(
FftSize: 1024,
HopSize: 256,
MelBins: 40,
CepstralCoefficients: 14,
MinFrequencyHz: 180.0,
MaxFrequencyHz: 15_000.0,
TableCount: 4,
HashBits: 14,
NearHashRadius: 4,
DenseStepSeconds: 0.040,
ProposalBudgetMultiplier: 8.0,
TemplateAugmentations:
[
new CepstralDegradationSetting("template-clean", 0.0, 0.0, 0),
new CepstralDegradationSetting("template-blur", 0.0, 0.0, 1),
new CepstralDegradationSetting("template-warp-light", 0.75, 1.25, 0),
new CepstralDegradationSetting("template-warp-blur", 0.75, 1.25, 1)
]);
}
internal sealed record BioacousticTrainingHypothesis(
string Id,
string Notes,
CepstralDecoderOptions Decoder);
internal sealed record CepstralRoundTripAnalysis(
int FrameCount,
int MelBins,
@ -1806,3 +2183,63 @@ internal sealed record CepstralWordObservation(
internal sealed record CepstralWordIndex(
IReadOnlyList<CepstralWordTemplate> Templates,
IReadOnlyDictionary<(int Table, int Hash), List<int>> Buckets);
[CultDocument("mimir.bioacoustic_training_result", "mimir.bioacoustic_training_result.v1")]
[MessagePackObject]
public sealed record BioacousticTrainingResult(
[property: Key(0)]
[property: CultName]
string ResultId,
[property: Key(1)] string RunId,
[property: Key(2)] string StartedAtUtc,
[property: Key(3)] string Toolchain,
[property: Key(4)] string HypothesisId,
[property: Key(5)] string HypothesisNotes,
[property: Key(6)] string DegradationName,
[property: Key(7)] int SampleRate,
[property: Key(8)] double Seconds,
[property: Key(9)] int ExpectedEvents,
[property: Key(10)] int ObservationCount,
[property: Key(11)] int CorrectAnchors,
[property: Key(12)] double Precision,
[property: Key(13)] double Recall,
[property: Key(14)] double Confidence,
[property: Key(15)] double TimingMeanAbsoluteErrorSamples,
[property: Key(16)] double DecodeMilliseconds,
[property: Key(17)] double RealtimeFactor,
[property: Key(18)] double IdentityScore,
[property: Key(19)] double TimingScore,
[property: Key(20)] double SpeedScore,
[property: Key(21)] double TotalScore,
[property: Key(22)] int RoundTripMelBins,
[property: Key(23)] int RoundTripCepstralCoefficients,
[property: Key(24)] int RoundTripFrameCount,
[property: Key(25)] double RoundTripRmsRatio,
[property: Key(26)] BioacousticTrainingDecoderSnapshot Decoder,
[property: Key(27)] BioacousticTrainingArtifact[] Artifacts,
[property: Key(28)] BioacousticTrainingObservation[] Observations);
[MessagePackObject]
public sealed record BioacousticTrainingDecoderSnapshot(
[property: Key(0)] int FftSize,
[property: Key(1)] int HopSize,
[property: Key(2)] int MelBins,
[property: Key(3)] int CepstralCoefficients,
[property: Key(4)] int TableCount,
[property: Key(5)] int HashBits,
[property: Key(6)] int NearHashRadius,
[property: Key(7)] double DenseStepSeconds,
[property: Key(8)] double ProposalBudgetMultiplier,
[property: Key(9)] string[] TemplateAugmentations);
[MessagePackObject]
public sealed record BioacousticTrainingArtifact(
[property: Key(0)] string Kind,
[property: Key(1)] string Uri,
[property: Key(2)] string ContentHash);
[MessagePackObject]
public sealed record BioacousticTrainingObservation(
[property: Key(0)] ulong EventIndex,
[property: Key(1)] double SampleOffset,
[property: Key(2)] double Confidence);

View file

@ -83,3 +83,4 @@
{"ts": "2026-05-23T23:30:55+01:00", "type": "decision", "status": "ok", "note": "Added source-level code algorithm map and linked Perfect Machine hypotheses research note."}
{"ts": "2026-05-23T23:45:11+01:00", "type": "decision", "status": "ok", "note": "Added Perfect Machine domain index plus 2026-05-23 architecture/optimization/reference study and sample implementation sketches for decoder, SIMD/GPU scoring, actuator, and native rings."}
{"ts": "2026-05-24T01:55:37+01:00", "type": "decision", "status": "ok", "note": "Completed corrected two-hour Perfect Machine study pass from 2026-05-23T23:54:42.9041588+01:00 through 2026-05-24T01:54:56.9747630+01:00; research/perfect-machine-study-2026-05-23 now indexes decoder architecture, acoustic field models, calibration sessions, Fensalir/native boundaries, distributed receivers, benchmarks, failure modes, low-level notes, and sample implementation sketches."}
{"ts": "2026-05-24T11:38:08+01:00", "type": "implementation", "status": "ok", "note": "Added bioacoustic indexed-cepstral training harness with CultLib/CultCache typed result receipts and WAV artifact directories. Latest run artifacts/bioacoustic-training/bioacoustic-20260524-103438 scored 20 hypothesis/degradation cases; identity often survives mel-cepstral warp/blur, but timing MAE under warp points next at a global clock/path hypothesis rather than larger brute-force matching."}

View file

@ -49,6 +49,10 @@ invariants:
- Faust/native DSP owns hot audio alignment, separation, spatialization, and
stems.
- Local secrets remain untracked.
- Mimir is a CultMesh app. Codebooks, schedules, calibration receipts, path
response, and compute-offload work need typed mesh-syncable state surfaces
so phones, microcontrollers, Raven, and Starfire can cooperate without
becoming independent timing authorities.
accepted_design:
- `Mimir.slnx` owns the app surface.
@ -121,7 +125,14 @@ accepted_design:
weighting. `Mimir.BufferSmoke --bioacoustic-cepstral-smoke` now proves the
non-brute-force receiver direction: degraded mel-cepstral calls are decoded
by an augmented MFCC/projection-hash word index before any expensive
confirmation. `MimirChirpBinTimeline` remains the controlled chirp-bin
confirmation. `Mimir.BufferSmoke --bioacoustic-train` now runs a
receipt-backed hypothesis panel over indexed cepstral decoder shapes and
writes typed CultCache results plus pre-warp/post-warp/reconstructed WAV
artifacts under `artifacts/bioacoustic-training/`. The latest receipt
`bioacoustic-20260524-103438` shows compact indexing is fastest on clean
audio, baseline indexing is more stable under light warp, highband indexing
can win blur/warp-blur cases, and timing error under domain warp remains
the pressure for a global clock/path hypothesis. `MimirChirpBinTimeline` remains the controlled chirp-bin
calibration/reference artifact with ASIO tooling and
`MimirChirpBinCalibrationModel`; keep those captures as response evidence,
not the active runtime sound. Reports and sync states expose delay in
@ -276,8 +287,8 @@ current_status:
next_action: >
Use the new bioacoustic timeline watermark as the active acoustic sync
direction. Cut the next implementation in this order: streaming indexed
log-mel/MFCC motif decoder state with a global clock hypothesis,
calibration-session command/replay proof that learns
log-mel/MFCC motif decoder state with a global clock/path hypothesis fed by
`--bioacoustic-train` receipts, calibration-session command/replay proof that learns
per-output/mic motif/formant response, fractional-delay/SRO actuator, then
native camera payload handles and Fensalir lowering. Keep chirp-bin
calibration captures as reference evidence, not the runtime target.