AI song detector
Check a song for free, then read what the score actually means. This page focuses on interpretation — the part most detectors leave out.
Upload Audio or Drag & Drop
Upload an audio file to receive a probabilistic analysis of characteristics associated with AI-generated music.
- MP3
- WAV
- M4A
- FLAC
- OGG
- AAC
- WebM
MP3, WAV, FLAC, AAC, M4A, OGG, WebM · max 25 MB · min 10 seconds · 30+ seconds recommended
Your audio never leaves your device. Decoding and analysis run entirely in this browser tab, and nothing is uploaded to a server. Your audio is processed only to perform this analysis, and your uploaded audio and temporary analysis data are automatically deleted after processing. No report links are created, and your analysis is never publicly accessible. Upload only audio you are authorised to process — analysis does not transfer ownership or publishing rights.
- Free
- Fast
- Secure
- No registration
Reading a detector score without over-claiming
The single most common mistake is treating a probability as a percentage of certainty. “72% AI” does not mean the track is 72% machine-made, and it does not mean there is a 72% chance the tool is right. It means: on the measurements taken, this recording sits 72% of the way along a scale whose calibration has not been externally validated.
The range matters more than the point estimate
We always show a plausible range alongside the number. When that range spans 57–77%, the honest reading is “leans generated, could easily be a heavily-mastered human track”. If a tool gives you a bare number with no interval, you are being sold precision that does not exist.
Confidence is a separate axis
Probability answers “which direction?”. Confidence answers “how much should you weight this at all?”. A 78% score at Low confidence carries less information than a 62% score at Moderate confidence, because the low-confidence run had worse audio, fewer usable segments or internal disagreement.
Inconclusive is a result, not an error
Roughly speaking, an inconclusive verdict appears when the sampled sections disagree, when encoding has stripped the detail the analysis needs, or when the score sits near the boundary. Forcing those cases into a yes/no answer is exactly how detectors generate false accusations. See accuracy and limitations for the specific triggers.
A practical reading scale
- Below 42% — nothing in the signal suggests generation. Weak evidence of human origin, not confirmation.
- 42–57% — no usable signal in either direction. Reported as inconclusive.
- 58–69% — some generated-audio characteristics. Worth a second look at provenance; worth nothing on its own.
- 70% and above — several measurements agree. Still an estimate, and still capable of being wrong about a loudness-maximised human master.
What to do with a leaning result
Go and gather non-acoustic evidence: project files, stems, dated drafts, the artist’s release history, distributor metadata, and — most usefully — a conversation. A detector score is the beginning of a question, not the end of one. The full process is set out in how to check if a song is AI-generated.