The music business has become a more litigious place with high-profile copyright lawsuits costing artists, labels and production houses millions of dollars. Songwriters and producers often come up with hooks that accidentally sound like existing songs, and rights holders have a hard time proving whether a melodic or rhythmic pattern was actually stolen. This requires deep technical analysis to separate actionable infringement from genre convention, something that standard audio recognition tools cannot do.
Forensic musicologist Dr. Ethan Lustig connects music theory, copyright law and technology. Dr. Lustig is a Ph.D. in music theory and has consulted on major federal copyright cases, including Bad Bunny “Safaera”, Jason Derulo “Savage Love”, Mary J. Blige, Anthropic, and Snoopy/Peanuts, where he has provided formal musicological analysis, expert testimony, and prior art research. He developed the online Music Copyright Checker to enable doing a preliminary analysis. The tool employs advanced algorithms to isolate pitch and rhythm and compute objective similarity percentages. We talked to Dr. Lustig about how algorithmic tools measure melodic overlap, the role of prior art in copyright defense, and the evolving science of forensic musicology.
Q: Federal courts have rolled out some new rules that are sending music copyright cases through the roof. When you’re actually comparing two songs side by side, why is it so hard to pin down whether one is infringing on the other?
Dr. Ethan Lustig: So when it comes to determining the extent of similarity between two melodies, there is no magic number of notes in common that suddenly make them “similar” vs. "different". The context is everything. For example, one important factor to keep in mind is that even if two melodies are very similar, they might not be very original–they might be patterns already found in other music over the centuries. So in that case, they are not considered protectable by copyright to begin with. For example, two melodies could both go up or down the scale, and therefore be identical in terms of their pitches, but Bach, Beethoven, and the rest of them all used scales. (In fact, the US Copyright Office even says that chromatic scales, arpeggios, etc. are not granted copyright protection.) This gets into the idea of “prior art”–prior, existing compositions from history that already used the pattern we are looking at. If a pattern is found in earlier songs, then it raises questions about whether it really is protectable by the plaintiff in the first place. Finally (and this is not what we musicologists handle, by the way, but rather the lawyer), in order to even have a case, plaintiffs need to demonstrate what’s called “access”–the idea that the artist who allegedly copied you would have reasonably heard and accessed your music in the first place. So it’s often considered a one-two punch of access and substantial similarity, but in reality it’s more like a one-two-three waltz of access, similarity, AND a shared pattern that cannot be explained away by prior art.
The other thing I'll say, is that while untrained listeners might perceive two songs to be super similar, that can often just come down to factors that aren't actually important from a music copyright perspective: the two songs might just be in a similar tempo, key, or instrumentation, none of which are as important for substantial similarity, as having a similar melodic pattern…only a trained forensic musicologist can capably filter through the noise and determine what similarities are, and aren't, meaningful from a forensic perspective.
Q: You’ve also got this online Music Copyright Checker that lets people test their own rhythms and melodies against existing tracks. Why does the similarity analysis have to treat rhythm and pitch as completely separate things?
Dr. Ethan Lustig: So first of all, brief definitions. Rhythm refers to where notes enter, over time. For example, if I tap out the rhythm of “Happy Birthday” on the table, even though I am just tapping and there are no specific pitches (e.g. A, B, C, etc.), you can still recognize that it’s the rhythm of “Happy Birthday”. By contrast, pitch refers to what is often colloquially called “the notes”–A, B, C, etc.–and to continue this analogy, you could theoretically perform “Happy Birthday” with a different rhythm (imagine a rhythmically impaired uncle at the birthday party) but still get the pitches correct. Now we can basically present the formula for “melody”, which is: pitch + rhythm (contour can also be important, but that's for a separate discussion). Just the rhythm of “Happy Birthday” tapped on the table isn’t really a proper rendition. And just the pitches of "Happy Birthday" performed without its characteristic rhythm isn’t, either. But once you put the two together, you have now presented the composition, “Happy Birthday”. So anyways, when we are listening to music, we are perceiving both pitch and rhythm at the same time. And in court, the question is usually about the two in combination as heard, not as separate scores. But nonetheless, research also demonstrates that the two things are separable: for example, there are some people who are great percussionists but are “tone deaf”–they simply can’t sing a tune to save their life–and other people who can sing beautifully in tune, but have no sense of rhythm. (I've seen this with my own students at the University.) Now, it happens that some melodies have a strong similarity in pitch but not in rhythm, or vice versa. Of course the strongest similarities involve a similarity of both pitch and rhythm (with an exact match being the strongest possible case). But pitch is measured in pitches (as pitch series), and rhythm is measured over time (for example in beat locations). And because they are two different domains, with two different measurements, and two different cognitive realities, they are worth measuring separately in many cases. Think about how a painting could be analyzed by its color content vs. its geometric shapes, for example. The colors might all be the same but the shapes are completely different; or vice versa.
Q: The Melody Comparison tool leans on the Levenshtein edit distance plus some key-finding algorithms to measure how close two melodies are. How does that math deal with things like extra notes or the fact that songs might be in different keys?
Dr. Ethan Lustig: Sure, so as you mentioned, my Melody Comparison tool uses a key-finding algorithm to determine the key of the user-inputted melodies. That's critical because, if Melody 1 is in C major, and Melody 2 is in G major, we can't make any meaningful comparison until we can get them into the same key. This is because the specific pitches, the key of a melody, this isn't what's important. You can hear a cover of "Yesterday" by The Beatles in C major, D major, E major…it's still a beautiful song and you probably won't even notice the difference. What determines melody is the specific relationships between the notes–which form what are called "scale degrees"–this is sort of like the DNA of the pitch aspect of a melody. And so what my Melody Comparison tool does, is determine what keys both of the input melodies are in, and then transpose them into scale degrees so that meaningful similarity comparisons can be made. In terms of the extra notes piece of your question, my algorithm averages between the note counts of the two input melodies; so if Melody 1 has 10 notes and Melody 2 has 5 notes, the average note count is 7.5 and the similar note count is divided by 7.5 to obtain a similarity percentage. It's very important to normalize for differences in note count between two patterns, because if you don't do this, you can just manipulate the data to say that any shorter melody which is a subset of a longer melody, is 100% similar, which is simply not true. The notes that are not sung are just as important as the notes that are sung. Rests (silences) are critical for the rhythmic identity of a melody.
Q: Even if two songs score really high on similarity, that doesn’t automatically mean there’s a copyright violation in court. How can “prior art” protect artists when the musical elements they’re using are already floating around out there?
Dr. Ethan Lustig: That's correct. You can have two melodies with a 100% identical shared stretch, in both pitch and rhythm, but if that same context occurs in a pre-existing composition (what's called "prior art"), musicologists should perform "filtration" to remove that segment from consideration. It then tends to be treated as no longer protectable. Prior art is so important that it's treated as required in an opening expert report (at least in the Ninth Circuit, which includes California, where the majority of high-profile music copyright lawsuits happen). In fact, multiple musicologists have had their report excluded, and the case severely hurt, by not doing prior art analysis in their report. Prior art can greatly protect artists because, working with a good musicologist, you can demonstrate that you didn't copy someone else's melody–because that melody also is found in another older composition. Therefore, it's part of a shared lexicon. Anyways, prior art is one of the messiest battlegrounds out there, so you really want to tread carefully and make sure to find a forensic musicologist who is a prior art specialist. I've actually even been subcontracted by other forensic musicologists for my prior art consulting.
Q: Automated tools are great for a fast first pass, but what pieces of context and substantial similarity still really need the trained ear of a forensic musicologist?
Dr. Ethan Lustig: My Melody Comparison and Rhythm Comparison tools are there to inspire people and let them get a head start with playing around with some of these comparisons. But you still really need a pro musicologist and that's because only a pro will be able to give you the peace of mind that they are correctly representing the notes, correctly analyzing the context; they know what type of similarities are and are not significant (prior art); and on top of all of that, you need a qualified testifying expert in a court of law if you want to have any weight in your case.
Q: What should artists, commercial producers, and media companies actually do while they’re still in production so they don’t get hit with expensive clearance fights after the track is already out?
Dr. Ethan Lustig: Basically, what ad agencies, music houses, labels, etc. should do is to hire a musicologist to conduct an assessment of their composition, to make sure there are no issues of significant similarity with other compositions out there. There's a saying in Spanish that "lo barato sale caro" that means something cheap at first, often ends up being more expensive in the long run. Companies often hire a composer to make music similar to the "temp track" (reference material) and then they try to risk it and not hire a musicologist; then sooner or later, they suffer for it and end up getting sued or have to settle out of court. And a lot of these payouts, even for instrumental music, are 6 figures. So what's better, paying 3-4 figures for a one-time clearance analysis from a musicologist, or paying 6-7 figures in a lawsuit that you could have avoided? And then you also have to remove your ad, that you already paid marketing spend, wrote, filmed, and edited, from all platforms?! What I offer in clearance letters is to identify if the composition is substantially similar not only to the temp track, but also to any pre-existing compositions that are out there; or if it's safe and "good to go". If there is a problem, I let the client know and even can suggest easy-to-fix changes that the composer can apply so that there are no longer infringement issues.
Q: What’s your take on the use of AI for music plagiarism detection? Can’t people just ask ChatGPT if the two melodies are similar?
Dr. Ethan Lustig: That's a great question. I've actually consulted for Anthropic as a forensic musicologist, and I've worked as an "AI data trainer", helping to align some of the very chatbots that are out today, which gave me a better understanding of how these systems work. As of today (we are currently in September 2026), AI is definitely very powerful for a lot of tasks. But it's still not able to provide effective forensic musicology analysis, for a few reasons. First of all, an AI chatbot is able to do an Internet search for information. It's able to perform reasoning. But it's not able to accurately extract musical parameters via "listening" to an audio file (much less a YouTube link). Once the user realizes this, then they might be tempted to input the melody themselves (for example, in music notation or as typed-out notes). The problem there is that the user may not even be writing the correct pitches, rhythms, and other context, and even if they do write it correctly, the AI might still misread it. (It can misread music notation or pitch text.) Clients frequently contact me saying, "I was speaking with this AI chatbot the other day, and it said my melody is substantially similar to this other artist's melody, and that I have a strong case." I have even seen clients show me drafted forensic reports written by chatbots! Of course the second point is that, any reporting from a chatbot is not admissible in court–you can't use an AI as your expert witness, and using an AI report instead of a human-written expert report is a good way to hurt your case. If you want to win your case in court, you will want a legitimate, proven expert musicologist, who has testified, been deposed, etc., and not a chatbot.
Figuring out whether two pieces of music truly overlap takes a careful mix of hard numbers and real musical judgment. Dr. Ethan Lustig shows how crucial it is to separate what’s actually protected by copyright from the basic musical building blocks that nobody owns. That distinction keeps creators safe while still leaving room for artistic freedom. When you look objectively at pitch, rhythm, and harmonic structure, artists and legal teams can catch potential problems early and walk into any dispute with solid, reliable evidence on their side.
Commercial music production is accelerating at the same time that digital libraries are expanding, and the combined effect is a rising need for precise copyright analysis and reliable clearance verification. Algorithmic comparison tools deliver quick results, but those results become far more accurate when they are examined by a forensic expert, giving both creators and rights holders a practical (and legally valid) form of protection. Specialized musicological expertise remains essential for preserving artistic integrity and intellectual property rights as the creative landscape continues to evolve.
To learn more, visit EthanLustig.com/Forensic-Musicology