AI Music
Detector V20.
Browser-only AI music detection with Quick Analysis, Reference Analysis, local learning, calibrated probabilities, Bayesian evidence fusion, Human / Hybrid / Fully AI classification, confidence intervals, and full forensic reporting.
Quick Or
Reference.
Use Quick Analysis for one song, or Reference Analysis to add up to six Human or AI references. Every reference is weighted equally at the user level; the detector decides influence based on quality, similarity, and evidence reliability.
Upload Main Song
This is the song being judged.
Reference Analysis supports up to six optional references. Each reference must be labeled Human or AI. If AI, choose the generator if known.
Simple
Assessment.
The basic AI / Human / Hybrid answer will appear here.
Calibration +
Probability.
Isotonic calibration, Platt scaling, beta-binomial uncertainty, Brier score, log loss, MCC, ECE, Dempster-Shafer, softmax, and hierarchical classification will appear here after analysis.
Reference
Evidence.
In Reference Analysis mode, uploaded references become part of the final scoring equation.
Explanation
Engine.
The detector will explain why it chose the final verdict and why alternatives were rejected.
Trust Each
Signal.
Each evidence family will receive a score and reliability rating.
How The Score
Was Built.
The decision replay will show the complete path from raw score to calibrated final probability.
Was It
Correct?
After a scan, tell the detector whether the result was right. The learning stays in this browser only.
Votes and calibration data will be stored locally in IndexedDB.
Self-Test +
Browser Check.
Self-test results will appear here.
Reports +
Data.
Run an analysis to enable report exports.
Simple Answer.
Deep Math.
V20 keeps the basic answer first while adding calibrated probability, uncertainty, reference consensus, and professional forensic transparency underneath.