High-frequency cutoff & artifacts
The spectrum shows where energy falls away. A sharp drop near 15–16 kHz can follow generation, an MP3 export, resampling, or a filter applied during mastering.
Inspect a song’s frequency spectrum, dynamics, rhythm, and metadata for patterns sometimes associated with AI-generated music. See the measurements behind every score.
Private audio analysis with a heuristic score
Analysis runs entirely in your browser. Your audio never leaves your device. The score summarizes clues from this file. It gives no calibrated probability of AI generation and cannot serve as forensic or legal proof.
Drop an audio file here
MP3, WAV, FLAC, M4A, or OGG · Up to 50 MB
A short browser-synthesized clip for exploring the tool.
Drop a supported recording or try the synthetic demo. Your browser decodes it locally.
Play the preview and follow it on the spectrogram. Read the four cards to see which measurements contributed to the score.
Copy the report with the measurements and caveats. For attribution, compare it with release credits, project files, or a documented source.
The spectrum shows where energy falls away. A sharp drop near 15–16 kHz can follow generation, an MP3 export, resampling, or a filter applied during mastering.
A low peak-to-RMS ratio and a narrow loudness spread suggest a consistently dense signal. Limiters and modern pop or EDM mastering can produce the same pattern.
Repeating onsets, close agreement between channels, and stable high-frequency phase can also come from drum machines, quantization, or centered mixing.
Generator names in software or comment fields are leads worth checking. Generic encoders describe how a file was packaged. Studio tags and ISRCs describe its production or catalog context. Any of these tags can be edited.
Choose based on the audio you can share, the measurements you need, and how many tracks you plan to review.
No. The score summarizes measurable patterns that can also appear in human recordings. A low score does not establish human authorship. For attribution, look for production records, credits, and original project files.
The detector reads, decodes, and analyzes audio in your browser. It does not upload or persist the audio or report. Copying a report places the text on your device’s clipboard.
Use an MP3, WAV, FLAC, M4A, or OGG file up to 50 MB. Decoding depends on browser codec support. The detector samples up to 60 seconds across long recordings and gives clips under one second less weight.
A cutoff shows a drop in high-frequency energy. Compression codecs, sample-rate conversion, recording equipment, and mastering filters can all cause it. Try an earlier lossless export if you have one.
Tags can be stripped during export. FFmpeg and Lavf describe how a file was encoded or packaged and do not raise the score. A generator tag raises the indicator, but tags can be edited.
The metadata panel can show a generator name if an available tag mentions it. The audio heuristics do not identify a model or isolate vocal stems, so hybrid recordings can get any score.
Look up the identifier for a specific recording.
Format a code using your assigned registrant details.
Find the UPC for a music release.
Start a song from an idea or separate an existing track into stems.