43 Meta Data
(Modified on 2026-08-11)
1,229 words
Metadata is data about other data. In the context of music research, “metadata” usually refers to data about the symbolic or audio data that is the main focus of an inquiry. In a similar vein, we might regard demographic or personality data gathered from human participants to be a form of metadata.
Metadata may or may not be relevant to a lot of research. If aspects of metadata are directly relevant to research questions—either as independent/dependent variables or covariates to control—then they aren’t really “metadata,” just data. Metadata may also be important as a sampling criteria—for example, sampling or grouping data based on year of publication.
Types of Metadata
Biographical and Historical
Much metadata is basic biographical/historical information, especially the names of composers or performers and the date of composition, recording, and/or commercial release. Generally, such data reflects the concerns of the music industry: for example, recognizing copyright-holders and “official artists” but rarely studio musicians or engineers. This reflects a particular narrow, biased view of musicking, to some extent. The positive flip side is that, because the music industry is highly concerned about this sort of data, it is relatively easy to find—often embedded directly into audio files—and reliable. Conversely, finding more obscure information about, e.g., recordings or scores, can be difficult. Biographical/historical metadata is generally uncontroversial—usually it is well-documented by “official” sources.
Due to the importance placed on composers/performers, much metadata might spend more time describing them. For example, metadata might include the dates a composer lived—e.g., “J.S. Bach, 1685–1750.” Sort of like metadata about the metadata.
Metadata may include plain-old errors, like typos, but rarely serious disagreements. Still, we must maintain a skeptical and concerned eye regarding construct validity. For example, Hey Jude is officially credited to John Lennon and Paul McCartney but by most accounts was written entirely by Paul. The official “Lennon and McCartney” data point is easy to find, but documenting who actually wrote each tune can be difficult to verify—and that’s for the Beatles, who are probably the most documented artists in history!
Dates
Musical pieces/performances/recordings have many dates associated with them. In rough chronological order, we might see dates like:
- Date of composer(s) birth.
- Date artist/band officially formed.
- Date of composition.
- Date of recording/production.
- Date of official release.
- Date of first appearance on Billboard Hot 100.
- Date of peak chart position.
- Date of remaster release.
Some of these dates are better defined and easier to reliably measure than others. But which “date” is the relevant date for your research?—that depends on your question! For recordings, the “date of release” is usually the easiest date to find and is most likely to be considered “the date” of the music. However, it is pretty common for, for example, a pop song released in December to “chart” the following year, and you might see “hit songs” dated in multiple years.
In most cases, “date” metadata is just a year. However, if we are studying something like the charting history of pop music, having more precise (monthly, or daily) date information might valuable.
Here’s a question: Let’s say we want to study how rock music evolved between, say, 1960 and 2010. If you sample a large body of popular recordings you will find that, for example, bands like the Rolling Stones and Pink Floyd released commercially successful albums in the 1960s, 1970s, 1980s, and 1990s. However, these bands are much more strongly associated with earlier decades. So, how should we count their later tracks? Should they be considered examples of “70s-style” even if released in the 1990s? Or should they be counted in their true date of release?
Watch out! Lots of commercial music is periodically remastered and released every few decades. If you access a recording of Hey Jude, it’s quite likely that the version you are hearing was remastered (and of course, digitized) decades after the original release. So, which date should it be counted in? If we are studying the composition, then probably 1968. But if we are studying the recording—maybe the later date is more representative?
Descriptive
Some metadata focuses on describing the music. Some descriptive metadata may be fairly uncontroversial and reliable: for example, track lengths (though different versions of recordings may be different lengths!). Other descriptive metadata may be more obviously debatable: things like genre or mood labels; music theory information like meter, tempo, form, or key; or the instrument lists. If you find such metadata, you should always question the provenance of such data. Was it extracted algorithmically? Labeled by (lay or expert) human participants? Pile-sorted from critical reviews (in magazines, blogs?)? Released in “official” record liner notes? In many cases, you either can’t find or can’t trust the metadata you need, in which case you’ll need to measure it yourself.
Errors
Metadata can and does have errors! Check for errors! In my paper on “tempo stability” (2024) we found multiple tracks in our Spotify data which were labeled with “hip hop” as the genre label, but with “release years” in the 1920s!
Formatting
One of the most common problems with metadata is inconsistency. Are they "Beatles, The", "Beatles", or "The Beatles"? "Blink 182" or "Blink-182"? Was the album "Metallica (the Black Album)" or just "Metallica"? Is the song "Sittin' on the dock of the bay" or "(Sittin on) the Dock of the Bay"?
It is very, very common for artist, album, and song-title data to have these kinds of inconsistencies. When you make your own data, take the time to do better! Pick a consistent style—for example, with “The” always before artist name—and stick to it. When working with other people’s data—carefully check for these problems. Experiment with doing things like:
- Converting all strings to lower case.
- Removing everything inside parentheses.
- Removing “The” at the beginning of titles.
To search for possible duplicates or near duplicates, a great strategy is to print out all your metadata (all the titles, or all the artist/album fields, etc.) sorted, and scan for duplicates by eye. Depending on your sampling goals, including duplicates in our data may or may not be desirable, but you definitely want to be aware of them if they are there!
Filenames
In computational work, one of the most important and ubiquitous forms of metadata is the file names that are used. File names can be ridiculously inconsistent, both between and within data sets. Many of the problems mentioned in the previous section (“Formatting”) are often found in file names. Watch out for the use of spaces vs hyphens vs underscores!—case, track numbers…etc.
Hey_Jude_1968.wavheyjude.wavHey-Jude.wav01_HeyJude.wavhey-jude-live-in-berlin.wav- etc.
Conversely, file names often have a lot of useful information in them. For example, you might search for file names that contain keywords like Live or Remaster, as operational definitions for those constructs.
Best Practices
When you prepare your own datasets, especially if you are going to publish them, take the time to make sure your metadata is consistent and correct. If you publish data, always include details about how the data was created, how metadata was created/found and attached to the data, and how your metadata was formatted. This can go, for example, in a git-repo README.md file. Make sure your directory and file names are consistent and logical!