Free tool
Audio metadata checker
Before you analyse the sound, read what the file says about itself. This reads the tag block inside an MP3, M4A, FLAC, Ogg or WAV — encoder strings, software fields, ISRC, titles and comments — and explains what each one is and is not worth. The file is parsed in your browser and never uploaded.
Short answer
Does an audio file's metadata show whether it was AI-generated?
Sometimes, and only when the file says so: some exports leave a generator name in an encoder or comment tag, which is written provenance and the strongest evidence a file can carry about itself. Tags are also editable and are stripped by most upload paths, so empty metadata means nothing, and neither a tag nor an acoustic score is proof.
What a detection result means →Drop an audio file here
MP3, M4A, FLAC, Ogg, Opus or WAV. Only the file header is read.
Why provenance beats acoustics — when it exists
Our own ordering of evidence puts documentation above listening, and listening above any score: see the triage order we recommend. A tag that names a generator, a distributor registration, a session file with a date — each of those describes how a recording came to exist. A detector score describes only how the waveform behaves, and infers the rest.
That is why this tool exists as a first step rather than a feature buried in the report. If the file volunteers its origin, no probability is needed. When it does not — which is most of the time, because uploads strip tags — the acoustic route is what remains, with the limits set out on the accuracy page and the measured variance on score stability.
What gets stripped, and where
Platform transcodes rewrite the container and usually discard software fields. Stem exports and WAV bounces often start with no tag block at all. Messaging apps re-encode aggressively. By the time a file reaches you second-hand, expect the metadata to be gone — which is exactly why its absence carries no information about AI involvement.
If you are the one publishing, the durable record is the one you write down: state involvement accurately at upload and keep your project files. That trail outlives every tag block and every score.
Questions about audio metadata
- Can metadata prove a song was made with AI?
- Only when the file itself says so, and even then it is a claim rather than proof. A tag naming a generator is a strong lead because it is written provenance, not an inference from sound. Tags can also be added, edited or stripped in seconds, so a clean file proves nothing either way.
- Why is my file's metadata empty?
- Because most upload and download paths strip it. Platform transcodes, WAV exports, stem bounces and messaging apps commonly discard tag blocks while keeping the audio intact. Empty metadata is the normal case, not a red flag.
- What does the encoder or software field tell me?
- How the file was compressed, and occasionally which application wrote it. Names like LAME, FFmpeg or iTunes appear on human and AI-generated material alike, because nearly everything is re-encoded somewhere on the way to a listener.
- Is my file uploaded when I check it?
- No. The first part of the file is read locally with the browser's file API and parsed in the page. Nothing is transmitted, logged or stored, which is the same architecture the detector itself uses.
- Should I check metadata before running detection?
- Yes. It costs one click and it sometimes answers the question outright. When it does not, acoustic analysis gives you a probability from the audio, which is separate, weaker evidence about a different thing.