Spaces:
Sleeping
Sleeping
File size: 3,276 Bytes
127b976 | 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 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 |
from __future__ import annotations
from typing import Any, Dict, List, Optional
def safe_decision_gate(
fake_probability: float,
attack_type: str = "unknown",
suspicious_segments: Optional[List[Dict[str, Any]]] = None,
quality_warnings: Optional[List[str]] = None,
ood_warning: bool = False,
real_threshold: float = 0.35,
fake_threshold: float = 0.85,
window_threshold: float = 0.75,
) -> Dict[str, Any]:
suspicious_segments = suspicious_segments or []
quality_warnings = quality_warnings or []
max_window_score = 0.0
if suspicious_segments:
max_window_score = max(float(x.get("fake_score", 0.0)) for x in suspicious_segments)
evidence_consistent = (
fake_probability >= fake_threshold
and attack_type not in {"bonafide", "unknown", ""}
and (not suspicious_segments or max_window_score >= window_threshold)
)
warnings = []
if quality_warnings:
warnings.append("Audio quality may affect reliability.")
if ood_warning:
warnings.append("Audio may be outside the model training domain.")
if fake_probability < real_threshold and attack_type == "bonafide":
decision = "likely real"
review_required = False
elif fake_probability < real_threshold:
decision = "likely real"
review_required = False
warnings.append("Fake probability is low, but the attack-type head is not fully aligned.")
elif fake_probability < fake_threshold:
decision = "suspicious / human review"
review_required = True
warnings.append("Fake probability is in the uncertain middle range.")
elif ood_warning or quality_warnings:
decision = "suspicious / human review"
review_required = True
warnings.append("High fake score, but reliability warnings are present.")
elif evidence_consistent:
decision = "likely fake"
review_required = False
else:
decision = "suspicious / human review"
review_required = True
warnings.append("High fake score, but evidence is not fully consistent.")
return {
"safe_decision": decision,
"review_required": review_required,
"warnings": warnings,
"thresholds": {
"real_threshold": real_threshold,
"fake_threshold": fake_threshold,
"window_threshold": window_threshold,
},
"evidence_consistent": evidence_consistent,
"max_window_fake_score": max_window_score,
}
def make_safe_report(report: Dict[str, Any], quality_warnings=None, ood_warning: bool = False) -> Dict[str, Any]:
gate = safe_decision_gate(
fake_probability=float(report.get("fake_probability", 0.0)),
attack_type=str(report.get("attack_type", "unknown")),
suspicious_segments=report.get("suspicious_segments", []),
quality_warnings=quality_warnings or [],
ood_warning=ood_warning,
)
out = dict(report)
out["raw_decision"] = report.get("decision", "unknown")
out["decision"] = gate["safe_decision"]
out["review_required"] = gate["review_required"]
out["decision_gate"] = gate
return out
|