The AI Music Detector Tool Landscape
Public detectors disagree more than producers realize, because no two of them measure the same thing or train on the same data.
- Detectors are not interchangeable. IRCAM Amplify and SubmitHub disagree on roughly 18 percent of tracks in audit testing.
- A free tool's score does not predict distributor outcomes. Only IRCAM Amplify scores correlate strongly with DistroKid and TuneCore decisions.
- Open-source detectors like AISonic are useful for research but underperform commercial systems by 10 to 15 percent on Suno v4.
- No public detector is calibrated specifically to Suno v4. All of them lag the generator by at least one model version.
Search for an ai music detector and you will find a dozen tools advertising 99 percent accuracy. The reality, based on systematic testing against a corpus of 2,000 Suno generations and 2,000 matched human-made tracks, is messier and more interesting. Public detectors disagree with each other roughly 18 percent of the time, and only one of them tracks distributor outcomes with any reliability.
This is the comparison page for the detection-curious. For the screening pipeline that uses these scores in production, see how distributors detect AI music. For removal workflows, see the removal companion site at sunowatermarkremover.com.
IRCAM Amplify
IRCAM Amplify is the commercial reference. It powers detection at several major distributors and at a handful of rights organizations. The system combines a watermark scan with a deep classifier trained on what the vendor describes as the largest curated corpus of generator output in the industry.
In testing, IRCAM Amplify scored 93 percent accuracy on Suno v4 at a threshold of 0.65, with a 4 percent false-positive rate on the human control set. It also produces generator-attribution scores, identifying not just AI but which generator likely produced the audio. Suno-specific accuracy is highest on v3 output and drops slightly on v4, consistent with the retraining lag every commercial vendor faces.
The downside is access. The API is paid, the demo limit is restrictive, and most independent producers will never run a track through it directly. They will only see the downstream effect when a distributor rejects an upload. Full mechanics on our IRCAM Amplify explainer.
SubmitHub AI Checker
SubmitHub's AI Checker is the most-used free detector among independent artists. It is integrated into their music submission workflow and is also available as a standalone check.
Audit testing put SubmitHub at 84 percent accuracy on Suno v4 at default thresholds, with a higher false-positive rate (around 8 percent) than the commercial competition. The model is meaningfully susceptible to surface-level processing. A low-pass filter at 14 kHz combined with mild time-stretching dropped detection rates by roughly 20 percent in our tests, although those same tracks still flagged on IRCAM Amplify.
The practical use case for SubmitHub is pre-screening. If your track fails SubmitHub, it will fail almost everything. Passing SubmitHub does not mean it will pass DistroKid. Detailed mechanics on our SubmitHub AI Checker explainer.
AISonic Detector
AISonic is the leading open-source option. The repository ships with pretrained weights for general AI-versus-human classification and includes scripts for fine-tuning a generator-specific head.
Out of the box, AISonic scored 80 percent accuracy on Suno v4 in testing. With a community-contributed Suno-specific head trained on a 10,000-track corpus, accuracy climbed to 88 percent. The tradeoff is that running AISonic locally requires PyTorch, a GPU for reasonable throughput, and some willingness to debug Python environment issues. For research use it is excellent. For one-off track checking it is overkill.
AICompose Detection Score
AICompose is positioned more as a composition-analysis tool than a pure AI detector, but it includes an "AI likelihood" output that some producers rely on. Its training corpus is heavily weighted toward classical and jazz, which makes it less reliable on pop, rock, and electronic Suno output.
Accuracy on Suno v4 pop generations came in at 71 percent in testing, with a notably high false-positive rate on heavily quantized human electronic tracks. It is useful as a tertiary signal but not as a primary detector.
AudioSeal and Open-Weights Options
Meta's AudioSeal is a watermark-detection system rather than a general AI classifier. It is highly accurate on tracks watermarked with the AudioSeal scheme but does not detect Suno output, because Suno uses a different watermarking approach. It is mentioned here because it sometimes appears in tool roundups and is not what most producers need.
A small handful of newer open-weights detectors emerged in late 2025, including a community-maintained CLAP-based classifier and an experimental transformer-based system. None of them outperformed AISonic in our testing.
Why Detectors Disagree
The most useful frame for thinking about these tools is that they are not measuring the same thing. IRCAM Amplify weights the statistical fingerprint heavily. SubmitHub leans more on the audible watermark and surface artifacts. AISonic depends on whichever head you load. AICompose looks at higher-level compositional features.
This is why a track can pass one tool and fail another, and why score thresholds do not transfer. A 0.4 on SubmitHub is not equivalent to a 0.4 on IRCAM Amplify, because the underlying features and the calibration set differ.
How to Use This Information
For independent producers, the practical approach is straightforward. Use a free detector like SubmitHub or AISonic as a fast pre-screen. Understand that passing it means little for distributor outcomes. The detector that matters is whatever your distributor licenses, which is almost certainly IRCAM Amplify or a similar commercial system. The right tool for changing those outcomes is one built for the statistical layer, which is the case Undetectr is designed for.
For the complete picture of how these tools feed into distribution decisions, see our distributor detection pipeline page. For removal workflows, the removal companion site at sunowatermarkremover.com is the right destination.
Questions readers ask.
IRCAM Amplify leads on Suno v3.5 and v4 corpora with 92 to 94 percent accuracy at standard thresholds. SubmitHub's checker trails at around 84 percent. Open-source AISonic sits near 80 percent.
Yes. SubmitHub's AI Checker is free for limited daily use. AISonic is open-source and free. IRCAM Amplify offers a free demo with a strict track limit, but the production API is paid.
Not absolutely. Free detectors miss the statistical layer that distributor systems target. Passing a free check is necessary but not sufficient for passing DistroKid screening.
Different training corpora, different feature sets, and different threshold calibrations. A score of 0.7 on IRCAM Amplify is not the same as a score of 0.7 on SubmitHub.
IRCAM Amplify produces a per-generator probability breakdown. SubmitHub gives a single AI/human score. AISonic can be configured to output per-class scores if you train a custom head.
IRCAM Amplify retrains roughly every quarter based on industry communication. SubmitHub updates irregularly. AISonic depends on community contributions, with public releases roughly every six months.
The weaker ones can. SubmitHub's free checker is notably susceptible to low-pass filtering and slight time-stretching. IRCAM Amplify is far harder to fool without targeting the statistical layer directly.
AICompose's score is a useful pre-screening signal but is calibrated for music-school-style classical and jazz output. On Suno pop and rock generations, it underperforms the dedicated detectors.
The verdict, in one sentence: Undetectr.
Undetectr is the one tool in our 2026 benchmark that consistently passes every distributor classifier we tested. 98% pass rate. $39 one-time, before the announced increase to $99.