Choose a source file
Use a file you own or have permission to analyze. Keep the original version and its source. Re-recording a speaker or repeatedly compressing a track changes the audio the detector receives.
Voice AI CheckerMusic betaSongs, singing & instrumentals
AI Song Detector
Check a music clip for signs of AI-generated vocals or instrumentals.
Upload a song you have permission to process, choose a segment, and review the two signals separately. A music check is a screening aid, not proof of how a track was made.
Free music betaUp to 30 secondsNo signup required
Checking a call, podcast or narration? Use the spoken audio detector.
Check this song
8–30 seconds per request. Up to 3 free music requests per UTC day, separate from voice checks. Pro and voice credits do not extend this beta limit. Failed requests may count toward the free limit.
Checking music service availability…
Only the selected segment is sent for third-party analysis. Read the privacy policy. Upload only audio you have the right to process.
Loading music checks…
Selected segment result
AI vocal score
AI instrumental score
These are model signals, not the probability that an entire song is AI-generated. The vocal score also reflects how much of this segment contains scorable vocals. “Not scored” means there was not enough relevant content to evaluate.
See segment-level signals
Times below are relative to the selected segment. Individual windows can be less reliable than the combined segment verdict.
| Time | AI vocal signal | AI instrumental signal |
|---|
Do not use this result alone to accuse an artist, reject a submission or settle a rights dispute. Check the original release, production records and the creator's explanation.
How to check whether a song was made with AI
Select a useful segment
Listen to the local preview, then enter the starting time. Choose 8–30 seconds containing the singing or instruments you want to examine. Avoid an entirely silent intro. The rest of the file is not uploaded for this check.
Read signals in context
Review the selected segment verdict, the vocal and instrumental scores, and any unscored windows. Compare the finding with the track's source and production history. A detector cannot replace those records.
AI songs, AI vocals and AI covers are different questions
A song can mix human and generated material. An AI singer may perform a human-written composition, a real singer may use a generated backing track, or a producer may use synthetic material in only one section. A single score cannot establish who wrote the lyrics or composed the melody.
This music tool separates vocal and instrumental signals when the selected segment contains enough relevant audio. It does not recognize a particular singer, establish consent for a voice clone, or identify the generator. Known difficult cases include AI backing tracks under live human vocals and AI background or choral singing.
Spoken introductions, podcast clips and calls belong in the AI audio detector for speech. A song's musical sections belong here. A low speech score should not be treated as evidence that a backing track was human-made.
Can I check Spotify, YouTube or SoundCloud links?
This version accepts local files, not streaming links. There is no link-import tab, and the tool does not retrieve a Spotify track or download a YouTube video. If you have a permitted source recording, upload that audio file directly.
The source file can be up to 20 MB and between 8 seconds and 10 minutes long. Playback and decoding support depend on your browser. You choose a starting point; the detector receives at most 30 seconds from there. A result cannot clear or flag the rest of the song, including a chorus or outro outside that window.
How accurate is AI song detection?
We have not established an independent music-detection accuracy benchmark for this website. Model confidence is not a measured accuracy rate for your file. High-production or heavily processed human recordings can be flagged incorrectly, and generated audio can be missed.
The summary scores describe different music signals. The AI vocal score weights scored vocal windows against the whole selected duration, so a brief vocal section can produce a low overall score even if that section is flagged. The instrumental score summarizes the scored instrumental windows. Neither is an ownership finding or a percentage of the song's creative work performed by AI.
If neither vocal nor instrumental content can be scored, the result is unclear. Do not interpret missing evidence as a human verdict. For a serious review, preserve the original and ask for recording sessions, stems or release provenance rather than repeatedly checking until a preferred answer appears.
What should I do with a flagged song?
Listeners can look for an artist disclosure or the original upload before sharing a claim. Artists can retain dated session files, raw takes and production notes to explain their process. Reviewers should use a consistent policy and let creators respond to flags; a short segment check cannot certify authorship or support a rights decision by itself.
To understand how we report real test outcomes without turning a few examples into an accuracy claim, see our real versus AI voice examples. Those examples test spoken voices, not songs, and are not a music benchmark.
AI song detector FAQ
Is this a free AI song detector online?
The music beta allows up to 3 free requests per UTC day, with 8–30 seconds analyzed per request. No signup is required. This allowance is separate from voice checks; Pro and purchased voice credits do not increase it. Service failures may count toward the free request limit.
Can I paste a Spotify, YouTube or SoundCloud link?
No. This version accepts local music files only. It does not download streaming links or bypass platform restrictions. Upload only audio you have the right to process.
Can it detect AI vocals and AI instrumentals separately?
The analysis returns separate vocal and instrumental signals when there is enough relevant content to score. AI backing tracks underneath a live human singer and background or choral AI vocals can be missed.
Can it identify Suno, Udio or the exact generator?
No. A result flags AI-related music signals in the selected segment; it does not identify a generator, singer, training source or copyright owner.
Why might a human-made song be flagged?
Heavily processed or highly produced recordings can be flagged incorrectly. A flag is a reason to review provenance, not proof that an artist used AI.
Is a low score proof that a song is human-made?
No. Short segments can miss AI sections elsewhere in a song, and the model can miss AI-generated content. No flag is not a certificate of human authorship.
What files and lengths can I check?
Choose an MP3, WAV, M4A, OGG or FLAC file up to 20 MB and between 8 seconds and 10 minutes long. Browser decoding support varies. Select a starting point; only the next 30 seconds, or the remaining audio if shorter, are checked.
What is the difference from the AI voice detector?
The voice detector screens spoken speech such as calls and narration. The song detector uses a separate music analysis to screen singing and instrumental content. Neither tool establishes identity or ownership.