Free tool
Batch AI music check
Drop in a set of tracks and get a probability range, confidence level and audio-quality note for each one, in a single table you can export as CSV. Built for the people who have to triage a submissions folder rather than judge one song. Every file is decoded and analysed in your browser — nothing is uploaded.
Short answer
Can you check several songs for AI generation at once?
Yes. This page queues multiple audio files and runs the same in-browser analysis on each, returning a probability range, a confidence level and a quality note per track, exportable as CSV. It is triage rather than proof: rows marked Inconclusive are genuinely unresolved, and low-bitrate or heavily limited files produce more of them.
What a detection result means →Drop several audio files here
MP3, M4A, FLAC, Ogg, Opus or WAV. Each track needs at least ten seconds of audio.
What a batch is good for
Nobody listens carefully to two hundred submissions. The realistic goal is to decide which twenty deserve a human ear, and that is a ranking problem rather than a classification problem. A table of probability ranges with confidence levels does that job honestly: sort by the top of the range, read the confidence column, and start with the tracks where the measurements were both high and reliable.
The failure mode to avoid is treating the column as a verdict per row. A single number near the boundary means the evidence was genuinely mixed, which is why we print a range and an explicit Inconclusive class instead of rounding. The reasoning behind that choice is set out on the accuracy page, and the measured run-to-run variance on score stability.
Why file quality dominates the results
Expect a folder of 128 kbps MP3s to return more inconclusive rows than the same music as original masters. Lossy encoding removes the high-frequency detail several measurements depend on, and heavy limiting flattens the dynamic behaviour of another. That is a statement about the files, not the music — our compression study quantifies the shift.
Before reading any score, it is worth checking what the files say about themselves. The metadata checker reads encoder strings and comment tags, which occasionally answer the question outright, and the spectrogram viewer shows in one picture whether a file has been through a codec.
Intake, not enforcement
If you are building a process around this — a label, a publisher, a sync library, a competition — the score belongs at the end of the process rather than the start. Ask for disclosure first, keep the answers, and use acoustic triage on the cases where the paperwork is thin. We set out that ordering in our intake guide for labels and sync, and the wording artists can use in the disclosure builder. For the mechanics of a large pass — file prep, triage order, where false positives cluster and how to document a row so it still makes sense months later — see how to screen a music catalog for AI tracks.
Questions about batch checking
- How many tracks can I check at once?
- As many as your browser can hold in memory, though the practical limit is a few dozen. Files are decoded and analysed one at a time and the audio for each is released as soon as its row is written, so a long queue works but takes time — expect a handful of seconds per track on a modern laptop.
- Are the files uploaded anywhere?
- No. Every file is decoded and analysed in your browser using the same engine as the single-track detector. Nothing is transmitted, logged or stored, and closing the tab discards the queue and the results. That is what makes the tool usable on unreleased catalogue.
- Is a batch result as reliable as a single scan?
- It is the identical analysis, so it is exactly as reliable and carries the same limits. Each row shows a probability range and a confidence level rather than a verdict, and rows marked Inconclusive should be treated as unresolved rather than rounded toward either answer.
- What should I do with a queue full of Inconclusive rows?
- Read it as a statement about the files rather than the tracks. Heavy limiting, low bitrates and short excerpts all remove the detail the measurements rely on, so a batch of platform-transcoded MP3s will produce more inconclusive rows than the same music as original masters. Re-run the ones that matter from the highest-quality source you have.
- Can I use the CSV as evidence in a takedown or a contract dispute?
- Use it as triage, not as proof. A CSV of probability ranges is a defensible way to decide which of two hundred submissions deserve a human listen. It is not a determination that any specific track was machine-made, and no acoustic method available today can be that.