Is This Song AI Generated?

A human vocal can sit over an AI-generated instrumental. A synthetic vocal can sit over a human-made arrangement. Both components may be generated, or neither may show a detectable signal.

The primary classification remains authoritative. The two component readings help explain that classification; they are not averaged into a manufactured overall probability.

Purpose-built models can learn patterns associated with known generation systems. Those patterns may support an AI-generated, not-AI or inconclusive classification.

Performance varies with unfamiliar generators, audio quality and later processing. A detector result is evidence about the submitted audio, not a philosophical judgment about whether an artist “used AI.”

A responsible AI song checker preserves uncertainty. If the system cannot return a usable classification, ANYANO shows inconclusive rather than silently turning uncertainty into a human result.

Work on music deepfake detection reports near-perfect scores against the generators a model was trained on, and a sharp drop against generators it has never heard. The same papers show that ordinary transformations — MP3 re-encoding, pitch shifting, time stretching, resampling — reduce measured performance, because they remove or move the very artefacts a detector relies on.

That is why no detector output, including this one, is evidence of authorship. A measurement can say a recording's acoustic profile resembles the profiles typical of synthesis pipelines. It cannot say who pressed which keys, whether a human performance sits underneath, or which service was used. Naming a specific generator would require a trained and validated attribution model, which this tool is not.

It also matters that authorship and copyright are decided by law and evidence, not by a spectrum plot. The U.S. Copyright Office's guidance on AI-assisted works turns on human contribution, not on how a file measures — so a detector result is at best a reason to ask a question, never an answer to one.

Modern generators can produce realistic singing and coherent arrangements. Human recordings can also contain aggressive tuning, synthesis, quantization, restoration and mastering that change the same kinds of patterns a detector may rely on.

Lossy compression can remove useful information. Remixing, resampling, pitch shifting, time stretching and re-recording can alter it further. Hybrid songs may also change origin from one section or layer to another.

For those reasons, detection should be treated as one technical signal alongside source files, credits, project history, disclosures and direct questions to the people involved.

A verse may be human-recorded while backing vocals are generated. An instrumental break may come from a generative system while the rest of the arrangement does not. Different layers can also be edited or replaced at different stages.

When segment-level results are available, the timeline shows where the system reported stronger or weaker vocal and instrumental indicators. It adds context; it does not prove the origin of a particular performer, stem or production decision.

The result cannot independently establish who created a song, who owns it, whether anyone acted dishonestly, whether AI was used intentionally, or the complete production history.

• Authorship and ownership. It cannot determine legal authorship, copyright ownership, contractual compliance or the identity of a creator.
• Exact generator. It does not establish whether Suno, Udio or another specific service produced the recording.
• Every component. A primary classification does not prove that every voice, instrument, stem or section shares the same origin.
• Intent or misconduct. It cannot show whether AI use was intentional, disclosed, permitted, dishonest or academically improper.
• Audio transformations. Mastering, compression, MP3 re-encoding, remixing, resampling, pitch shifting, time stretching and re-recording can affect the available signal.
• Difficult inputs. Low-quality sources, screen recordings, short clips, hybrid production and newer generators can produce false positives, false negatives or inconclusive results.

Your audio is securely submitted to ANYANO’s backend to perform the analysis you request. ANYANO holds it in memory for that request and does not write it to permanent media storage or add it to your music library.

Detector uploads are not used by ANYANO to train its Sound music-generation system or any other model. The specialist detection system processes the file to return the result; its handling is governed by its own terms. ANYANO records limited usage data without the filename or audio so it can enforce fair-use and abuse limits.

No account is required. A completed analysis includes the primary classification, separate vocal and instrumental indicators, reported confidence, segment information when available, and a downloadable PDF report.

You do not have to add the track to an ANYANO library. Free analysis is protected by fair-use, rate and concurrency limits, so availability is not presented as unlimited.

Use a direct export when possible. A higher-quality source generally gives the system a cleaner input than audio that has been repeatedly compressed, screen-recorded or passed through messaging apps.

WebM is not accepted. If your file uses another format, export one of the supported types before analysis.