What Does It Take to Identify a Tune?
Everyone has had the experience of hearing a few notes of a song and almost immediately identifying that tune. This website, through project Skiptune, attempts to answer the question, “What makes a tune unique?” Every melody must have some series of notes that does not appear in any other tune or melody–otherwise it would be a repeat of that other tune. Another way of posing the question is, “What’s the minimum series of notes in a tune needed to identify that tune as different from all other tunes?”
It’s obvious that you need at least two notes. What’s surprising is that after 66,000 melodies have been entered, roughly two percent of the tunes in the database only need two consecutive notes to identify them. We have found around 1,300 tunes out of 66,000 that can be distinguished from all other tunes by a pattern of just two consecutive notes. For instance, The House of the Rising Sun, a familiar song to most people, has such a pair. This version is taken from the songbook, “80 Years of Popular Music–The Sixties” by Warner Bros. Publications:
In the snippet of this tune, the syllable “ren” (in “children”) is sung to a dotted quarter note tied to another quarter note tied to an eighth note in a triplet, all in a D pitch. The word “not” is sung to an eighth note D that is up an octave and also in that same triplet. That pair of notes cannot be found in any other tune in our database, no matter what key it is in. In Skiptune’s shorthand parlance, that pattern is represented by the tuple [12, 0.118], which is a much shorter way of saying “the second note in the series is an octave above the preceding tied notes, and is 11.8 percent of the length of the set of tied notes.”
What Good Is All This?
Composers shouldn’t think too much–it interferes with their plagiarism.
While there are some strong academic reasons to care about this work, especially with respect to how composers use patterns to create songs, there are some practical uses as well. Here are a few examples:
- Generating Music. Artificial intelligence (AI) requires data to train on. The Skiptune project treats music, specifically melodies, as just data. A lot of data is needed to train an AI model, and we are still entering tunes into the database to achieve the critical mass necessary to train an AI model. The goal is to use AI to create new melodies that build on the amazing human achievement that is music. Unlike most large language models that can result in plagiarized generated results, the AI model we are creating will be guaranteed to produce a tune that has not been written yet. Some patterns of notes are used over and over again by many composers. These patterns may be codified in a way that will allow computer-generated music that sounds more human than attempts so far.
- Assisting Composers. Composers are like novelists or other writers in that they occasionally find themselves having written into a corner. They may have written a partial melody only to find that they have nowhere to go. That may be because they have written beyond where other composers have gone in a particular melodic line. The Skiptune database will eventually allow composers to enter their melody one note at a time and find what paths other composers had taken from each point (note) of the melody. By backing up to just before his/her tune is unique–that is, to the point where no one else had written those notes before–the composer can see where other composers had gone with the tune and either choose a path already taken by another composer or select a different branch no one has gone down before. Experimenting a bit, the composer can avoid writing him/herself into a corner.
- Ensuring a Song is Unique. By comparing a newly composed melody against those in the database, one can learn whether a new composition is a new creation or not. The other side of the coin is that the Skiptune database could provide litigants with hard data to argue their cases in court when there is contention over who wrote a particular melody first. It may be that the notes or major parts of the melody have already been written in the distant past in some obscure song in the public domain.
- Identifying Musical Works. It will be possible to use the Skiptune database to examine a particular historical musical piece whose author is in question and determine if the patterns used are similar enough to a particular composer’s known patterns, or to his/her era, to identify the true composer, or at least eliminate specific composers.
Where Do I Start?
The Skiptune project is complex and can be overwhelming at first. We suggest you start by looking at the FAQ page, followed by the Music Metrics page, using the Definitions page as needed to understand the terms used on this website. Once you have a feel for the way we categorize and compare musical patterns, you should look at the Example page to see the metrics in action on a simple tune. Follow that with Surprisingly Rare Patterns, which lists those patterns that would seem to be common, but are not. The rest of the website can be explored serendipitously.
© Skiptune 2014
Patent Notice: The algorithm we use to determine the uniqueness of a tune (identifying the sequence of notes that uniquely identifies a tune from all other tunes in the database) is protected under U.S. Patent 9,263,013 B2 and U.S. Patent 9,454,948 B2.