About & Methodology · Last verified 2026-11-22

About AIS Music Detector: How Screening Works & Known Limits

AI audio detection is statistical screening, not forensic proof of authorship. This page documents who runs AIS Music Detector, how 10-second spectral screening and conditional stem scoring work, which generators are covered, and where false positives happen.

Who builds AIS Music Detector

AIS Music Detector (aismusicdetector.com) is an independent audio-tooling project built for two audiences: listeners who want transparent segment-level context rather than a single black-box score, and independent artists diagnosing why a human-recorded master triggered automated screening on distributors such as DistroKid or TuneCore.

Our editorial and engineering boundary is strict: we do not build AI music generators, audio downloaders, or audio-editing tools meant to mask AI artifacts or evade distributor review. We do not provide legal advice or legal representation.

How spectral screening and three-tier verdicts work

Binary labels routinely misclassify heavily produced human tracks. Every scan maps raw acoustic probabilities into one of three documented zones:

Likely Human (<30%)
Unclear / Mixed (30%–70%)
Likely AI (>70%)
  • Analysis window: Up to the first 5 minutes (300 seconds) per track, minimum 10 seconds. Longer tracks are accepted and screened from 00:00 to 05:00 with coverage labeled directly on the report.
  • 10-second segment timeline: Spectral consistency and vocoder-like artifacts are evaluated in 10-second windows (up to 30 segments) so you can inspect peak intervals rather than trusting one opaque number.
  • Conditional vocal / instrumental stems: Vocal and instrumental bars appear when the detection engine triggers stem separation. When the full-mix signal is already conclusive or single-stem, the report displays Full mix honestly instead of inventing stem scores.

Supported and unsupported generators & audio conditions

Source / Generator familyCoverage statusNotes & practical boundary
Suno (v3.5 / v4 / v4.5)MonitoringFull-mix vocal and acoustic prompts evaluated in 10-second windows during pre-launch calibration.
Udio (v1 / v1.5)MonitoringEvaluated across pop, electronic, and acoustic outputs; lossless exports provide cleaner spectral signal than low-bitrate rips.
Hybrid human + AI stemsPartialTracks mixing human vocals with generated backing (or vice versa) frequently land in Unclear / Mixed; stem breakdown helps inspect vocal vs instrumental layers.
DRM-protected 30s streaming previews (Spotify, Apple Music, TikTok)UnsupportedBlocked before scoring without consuming a daily check. Short low-bitrate previews distort spectral scoring; upload the full file or use a public YouTube link instead.
Clips shorter than 10 seconds or non-music speechUnsupportedTracks under 10 seconds are rejected without using a check. Screening is tuned for music tracks, not spoken-word deepfake detection.

When results can be wrong (false-positive limits)

Audio conditionWhy it shifts scoresTypical score zoneHow to verify
Heavy Auto-Tune or pitch correctionStrong formant locking and robotic pitch steps resemble synthetic vocal harmonics.Unclear / Mixed (35%–65%)Keep dry vocal takes and DAW pitch-plugin chains alongside your bounce.
Brickwall limiting, AI mastering, or heavy denoisingAggressive spectral smoothing (for example iZotope RX or automated mastering) removes natural room noise.Unclear / Mixed (30%–60%)Compare the unmastered premaster mix against the final loudness-limited master.
Low-bitrate MP3 transcodes or short previewsRepeated lossy compression smears high-frequency transients above 14 kHz.Unclear / Mixed or inflated AI signalAlways scan the original lossless WAV/FLAC export rather than a compressed rip.
Clips shorter than 10 secondsInsufficient phrasing and transition context makes statistical estimation unreliable.Blocked before scoring (no quota used)Upload at least 10 seconds, ideally the full song up to 5 minutes.
Pure instrumentals and quantized electronic musicGrid-locked synths and drum machines share spectral regularity with generative loops.Unclear / Mixed (30%–70%)Retain MIDI clips, synth presets, and multi-track stem exports as creation records.

Internal benchmark calibration structure

We do not publish unverified single-number accuracy headlines. Our internal calibration set tracks false-positive rates on verified studio recordings, heavily tuned human vocals, instrumentals, and AI generations separately before publishing real figures:

Benchmark groupLikely HumanUnclear / MixedLikely AIStatus
Verified acoustic & studio human tracksCalibration pendingCalibration pendingCalibration pendingPre-launch set
Human tracks with heavy Auto-Tune / masteringCalibration pendingCalibration pendingCalibration pendingPre-launch set
Verified AI tracks (Suno / Udio generations)Calibration pendingCalibration pendingCalibration pendingPre-launch set

Platform and distributor review channels

No major streaming store or distributor accepts third-party AI detector screenshots as proof of authorship. When a release is flagged, reviewers expect primary DAW creation evidence such as session screenshots, dry stems, MIDI tracks, and commercial sample licenses.

Compare platform summaries on our main AI music detector page, read dedicated appeal guides for DistroKid and TuneCore, or review our Privacy Policy and Terms of Service.

Contact

For partnerships, media, or API attribution inquiries, email hello@aismusicdetector.com. For technical support, privacy requests, or rights-holder report takedowns, email support@aismusicdetector.com.

Frequently asked questions

Why does the timeline use 10-second windows?

Our detection backend evaluates spectral features in 10-second segments up to the first 5 minutes (300 seconds), giving up to 30 readable checkpoints per full track without distorting short phrasing.

When does vocal and instrumental stem breakdown appear?

Stem separation is conditional. When the detection engine triggers vocal/instrumental analysis, both bars are displayed. Otherwise we show Full mix honestly rather than inventing stem numbers.

Are the 30% and 70% tier cutoffs final?

They are our documented initial calibration thresholds (<30% Likely Human, 30%–70% Unclear / Mixed, >70% Likely AI) and are tracked publicly as our internal benchmark set expands.

Can AIS Music Detector prove authorship or override distributor decisions?

No. Acoustic screening is a probabilistic diagnostic aid. Distributors review DAW session files, stems, MIDI data, and licensing records rather than third-party detector scores.