Hugging Face's logo

TheHappyDrone
/
windflow-wind

Audio Classification
English
yamnet
wind
logistic-regression
noncommercial
Model card Files
xet
 Community
windflow-wind
File size: 2,083 Bytes
8357920
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
{
  "split": "ESC-50 folds: train 1\u20133; validation 4; test 5. Cross-fold source IDs purged from earlier splits.",
  "purged_train_source_ids": [],
  "purged_validation_source_ids": [
    "209698",
    "234879"
  ],
  "training_recordings": 1200,
  "validation_recordings": 398,
  "test_recording_count": 400,
  "threshold": 0.7500000000000002,
  "C": 0.001,
  "feature_type": "embedding",
  "selection": "Embedding, class-score and combined heads compared on validation only; highest F2 with validation FPR <= 3%.",
  "validation_windows": {
    "count": 1194,
    "positives": 24,
    "tp": 22,
    "fp": 26,
    "tn": 1144,
    "fn": 2,
    "precision": 0.4583333333333333,
    "recall": 0.9166666666666666,
    "f1": 0.6111111111111112,
    "fpr": 0.022222222222222223,
    "average_precision": 0.672848373136124,
    "roc_auc": 0.9916666666666667
  },
  "test_windows": {
    "count": 1200,
    "positives": 24,
    "tp": 14,
    "fp": 28,
    "tn": 1148,
    "fn": 10,
    "precision": 0.3333333333333333,
    "recall": 0.5833333333333334,
    "f1": 0.42424242424242425,
    "fpr": 0.023809523809523808,
    "average_precision": 0.44294890885194277,
    "roc_auc": 0.9736748866213152
  },
  "test_recordings": {
    "count": 400,
    "positives": 8,
    "tp": 4,
    "fp": 8,
    "tn": 384,
    "fn": 4,
    "precision": 0.3333333333333333,
    "recall": 0.5,
    "f1": 0.4,
    "fpr": 0.02040816326530612,
    "average_precision": 0.5041808490821649,
    "roc_auc": 0.9795918367346939
  },
  "limitations": [
    "Only 8 independent wind recordings in held-out fold; window samples are correlated.",
    "Labels denote the dominant event; non-wind recordings may contain incidental wind.",
    "Field recordings are not game/movie mixes; performance on capture-card audio remains unmeasured.",
    "Pretrained YAMNet was trained on AudioSet; upstream overlap with ESC-50 cannot be ruled out.",
    "Scores are classifier outputs, not a calibrated probability of real-world wind.",
    "ESC-50 is CC BY-NC 3.0; this experimental trained head is for noncommercial use."
  ]
}