Drop your Suno or Udio track. Whisper AI transcribes the lyrics in your browser, then we screen them against millions of known songs in lrclib.net to surface potential matches. Free, browser-only, no upload — catch accidental lyric regurgitation before YouTube's ContentID does.
Whisper transcription · lrclib.net database · Jaccard similarity · 1-2 min turnaround
Drag any MP3, WAV, M4A, OGG, FLAC, or AAC. The Web Audio API decodes it locally — nothing leaves your device. Works on any AI generator: Suno, Udio, Mureka, MusicGen, ElevenLabs.
Suno v5 outputs are the most common case but the tool is generator-agnostic.
@xenova/transformers Whisper Tiny.en runs in a Web Worker with multi-threaded WASM. The audio never uploads. Transcription takes 60-90 seconds for a typical 3-minute track. The model downloads once (39 MB), then is cached for every future visit.
Cross-origin isolation enables 4-8 thread parallelism. Without it, transcription would be 4-8x slower.
The tool splits transcribed lyrics into 4-word phrases, drops common ones ("I love you", "baby baby", short stopword runs), and ranks the rest by uniqueness. The top 5-7 phrases are the most likely lyric fingerprints — if Suno regurgitated something from its training data, it almost always shows up here.
Phrases from the chorus and bridge tend to be the most distinctive.
lrclib.net is a free public database of millions of synced lyric files for commercial music. Each distinctive phrase is queried via /api/search, returning up to 5 candidate songs. Most phrases return zero matches (good — that's the originality signal). Phrases that hit are the suspicious ones.
lrclib's privacy policy doesn't log queries with personal data. Searches are text-only — your audio never leaves your browser.
For each candidate match, the full lrclib lyrics are pulled and compared to your transcribed lyrics using Jaccard similarity on 4-word n-grams. 100% = identical. 50%+ = very likely a cover or major lyric copy. 15-30% = common phrasing (usually fine). Below 15% is dropped as noise.
The threshold is conservative on purpose — false positives are worse than missed soft matches.
The top 5 potential matches are displayed: track title, artist name, similarity score, and a snippet of the matching lines. You can click through to lrclib for the full lyric file. Clean result (no matches above 15%) is a green pass — your lyrics don't appear to overlap with anything in lrclib's database.
Always sanity-check matches manually. Common phrasing ("give me one more night") triggers false positives across multiple unrelated songs.
AI music generators are under active legal pressure for training on copyrighted material. While the courts work it out, distribution platforms have started flagging and removing AI tracks that match training-set songs. Catching matches before release is dramatically cheaper than fighting takedowns afterward.
Sony, Universal, and Warner sued Suno in June 2024 alleging unauthorized use of copyrighted recordings to train the model. The complaint cites specific Suno outputs that closely match charting hits. Lawsuit ongoing in 2026.
ContentID is increasingly aggressive about AI-generated music that contains lyric fragments matching commercial recordings. A flagged video has its monetization redirected to the rights holder and may be muted or removed entirely.
Spotify and Apple Music both reserve the right to remove tracks flagged as derivative or infringing, even after distribution. DistroKid and TuneCore have begun screening AI-generated tracks before they reach DSPs in some cases.
TikTok, Instagram Reels, and YouTube Shorts apply rights-holder-side audio matching. A Suno song that contains a verbatim lyric from a known track will get flagged, lose monetization, or be muted.
Honest limits. This is a lyric screening tool — useful, but not exhaustive. Read this before relying on the results for commercial release decisions.
We don't compare melody contours. A Suno song that copies a famous hook melody but has different lyrics will pass the lyric check. Melodic-similarity detection requires a melody database we don't currently maintain.
We don't compute audio fingerprints. Tools like AudibleMagic and Spotify's internal ContentID match acoustic features. Suno output rarely produces an exact recording match (the audio is genuinely synthesized) so this matters less, but we can't guarantee a Spotify ContentID will pass.
Drum patterns, chord progressions, instrumental arrangements — none of this is screened. Suno's audio is genuinely synthesized but its arrangement choices reflect the training data. Stylistic similarity to a specific producer is not legally infringing in most jurisdictions, but it can still trigger fan callouts.
lrclib indexes songs as their fan-submitted lyrics arrive. Songs released in the last 30 days may not be there yet. If Suno regurgitated a chart-topping hit from the last month, our check might miss it.
Whisper Tiny.en is English-only. If your Suno song is in Spanish, French, Hindi, etc., transcription quality drops sharply, and lrclib's non-English coverage is thinner. The tool will run but the result is less reliable for non-English material.
Phrases like "I'll be there for you" or "under the moonlight" appear in hundreds of unrelated songs. The tool de-emphasizes these but you should still read each match carefully. A 20% similarity score on common phrasing is usually fine; 50%+ on a distinctive phrase is the real signal.
Important
This tool is a screening aid, not legal advice. A clean result reduces but does not eliminate the chance of an infringement issue. For songs you plan to monetize commercially or distribute widely, consult a music rights lawyer if you have any uncertainty.
Yes — and there's documented evidence. Suno was trained on a large corpus of copyrighted music including the actual lyrics. The model occasionally regurgitates near-verbatim lines from training data. The RIAA's 2024 lawsuit against Suno cited specific examples. Always screen your output before release if you plan to distribute commercially.
No. This is a screening tool, not legal advice. We check lyrics against lrclib.net's public database — broad but not exhaustive. We don't check melody similarity, audio fingerprinting, or production techniques. A clean result here means the lyrics don't match anything in lrclib's catalog, but doesn't guarantee originality. For serious commercial releases, consult a music rights lawyer.
Suno claims fair use. The RIAA, Sony Music, Universal Music, and Warner Music disagree and filed lawsuits in 2024. The legal status of AI-generated music derived from copyrighted training data is unsettled in 2026. While that plays out, distribution platforms (DistroKid, TuneCore, Spotify, YouTube ContentID) increasingly flag and remove AI-generated tracks that match training-set songs. Screening before release reduces the risk of takedowns.
lrclib is a free public lyrics database with synced LRC files for millions of commercial songs across all major genres and decades. It covers most charting tracks since the 1950s, indie releases that have any kind of fan documentation, and a long tail of regional / non-English music. Coverage isn't 100% — newly released or extremely obscure tracks may not be there.
We split both your transcribed lyrics and each candidate's full lyrics into overlapping 4-word phrases (n-grams). The Jaccard similarity is the size of the intersection divided by the size of the union. A score of 100% means the lyrics are identical. 50% suggests significant overlap (a cover, a remix, or accidental verbatim copying). 15-30% usually means common phrases (e.g. 'I love you', 'one more night') rather than real copying. Below 15% we don't flag.
Lyrics carry their own copyright separate from sound recordings. Even if the music is genuinely AI-composed, regurgitating someone else's lyrics — even fragments — can trigger DMCA takedowns and ContentID strikes. YouTube's ContentID is particularly aggressive about lyric matches. Catching lyric copies before release is much cheaper than fighting a strike afterward.
No. The audio file stays on your device. Whisper transcription runs in a Web Worker in your browser. The only network calls are to lrclib.net's public lyrics search API, which receives 4-word phrases (text only, not your audio). lrclib's privacy policy does not log queries with personal data.
MP3, WAV, M4A, OGG, FLAC, AAC — anything the Web Audio API can decode. This covers every export format from Suno, Udio, Mureka, ElevenLabs Music, and standard DAW exports.
Transcription is the slow step — 60-90 seconds for a 3-minute song with multi-threaded WASM. Lyric search is fast (200-500ms total across 5-7 phrases). Total turnaround: about 1-2 minutes from drop to results. Whisper model download adds a one-time 30 seconds on first use; cached after that.
First, look at the match: if the similarity is below 30% it's usually common phrasing, not infringement. If above 30% with multiple matching lines, regenerate the song in Suno with a modified prompt that explicitly avoids the offending lyrics. Don't use the song commercially without consulting a music rights lawyer if a high-similarity match comes from a charting track.
Partially. With no vocals to transcribe, the lyric check returns nothing — which is a clean pass on lyrics but doesn't tell you about melodic or production similarities. For fully instrumental Suno output, the lyric route can't help; you'd need audio fingerprinting (ContentID) which is a server-side service we don't currently offer in-browser.
Yes. The tool doesn't care which AI generator made the track — it transcribes and screens any audio. Udio, Mureka, MusicGen, and ElevenLabs Music tracks all work the same way. The risk profile is similar: any model trained on copyrighted music can occasionally regurgitate from its training set.
Complete the workflow
AI Lyric Video Maker
Once you've cleared the originality check, make a TikTok-ready lyric video from your Suno track.
Spotify Canvas Maker
Auto 8-second Spotify Canvas from any AI track. 9:16, ≤30 MB, audio-reactive visuals.
Free AI Mastering
Master your Suno or Udio track for streaming after the originality check passes.
Suno Audio Fixer
Remove the metallic AI sheen from Suno output before mastering.
Suno Prompt Guide
Write better prompts to reduce the chance of Suno regurgitating training data.
Suno AI Prompts
50+ copy-paste Suno prompts that produce more original output.