Is This Song AI-Generated?
Upload an audio file and find out if it was made by a human or generated by Suno, Udio or another AI music tool.
Drop your audio here
or click to upload — mp3, wav, m4a, flac, ogg up to 20 MB
Free up to 2 checks per hour per IP.
Why Use Our AI Music Detector?
SOTA Open Model
Powered by SONICS (ICLR 2025) — F1 0.97 on Suno & Udio detection.
Fast Inference
Get a verdict in 5–10 seconds. Audio is processed and discarded immediately.
Genre-Agnostic
Works on pop, rock, electronic, classical and more — vocal or instrumental.
Free Forever
No sign-up. No payment. Just drop a file and get a result.
Privacy-First
Your audio is never stored. It's analyzed in memory and immediately discarded.
Honest Confidence
We show the actual probability percentage, not a binary yes/no — you decide.
How AI Music Detection Works
- 01
Upload Audio
Drop an mp3, wav, m4a, flac or ogg file — up to 20 MB.
- 02
AI Analyzes
SONICS computes a mel-spectrogram and classifies spectro-temporal patterns specific to AI music generators (vocoder artifacts, unnatural micro-variation).
- 03
See the Verdict
Get a probability score (0–100%) plus a binary verdict. Files are processed in memory and never stored.
Frequently Asked Questions
Which AI music generators can it detect?+
The current model (SONICS-α-120s, ICLR 2025) was trained primarily on Suno (≤ v4) and Udio (≤ v1.5) tracks from 2024. Newer versions, MusicGen, Stable Audio and Riffusion may not be reliably caught — we will update the model as new training data becomes available.
How accurate is it?+
On its own benchmark SONICS-α-120s reports F1 = 0.97. On real-world tracks accuracy is lower — heavy mp3 compression, pitch-shifting, mixing or cover-style edits all reduce reliability. Treat the percentage as a confidence indicator, not a final answer.
Is my audio stored?+
No. Files are received, analyzed in memory and immediately discarded. We don't keep copies, log content or use uploads for training.
Why is there a rate limit?+
Each analysis uses a real ML model on our server. To keep this tool free for everyone, we limit each IP to 2 checks per hour. Need higher volume? Email us.
Can I detect AI vocals on a real instrumental?+
Not yet — the current model classifies the whole track, not stems. Vocal-only detection (singing voice deepfake detection) is a different task and we're tracking research in that area.
Get the Genre AI App
Detect AI music, identify genres, and explore your taste — right on your phone.