HOW IT WAS DONE
The first step was to identify the defining element of a song, which Thompson and Daniels found was the musical composition. In order to measure the similarities, they used datasets created by the Music Genome Project. This project scores songs on over 400 attributes, which is then narrowed down to 8 data points by Echonest, a Spotify-owned music intelligence company and data platform.
The data points used were as follows: Acousticness • Danceability • Energy •Instrumentalness • Liveness • Loudness • Speechiness • Valence
The final statistics are calculated based on the average distance between data points on a graph.
TRANSPARENCY / OPENNESS
Overall, the authors did a good job at displaying their data and explaining the process behind their research. They have clear sources for research, even though a few of them are confidential or are trapped behind paywalls.
With a quick internet search, Echonest seems easily accessible, which is the base of this project. They are very open to the public about their research and tools, and it looks like their analyzing tool is free as long as you sign up as a developer on their site (https://github.com/echonest/en_analyzer).
“THE MUSICAL DIVERSITY OF POP SONGS” is very clear with its method, and provides enough information to allow others to create similar projects.
Methodology
Matt Daniels and Andrew Thompson do a great job on showing where they got their research. Everything mostly leads back to the company EchoNest. Although, how reliable is EchoNest?
EchoNest is a “music intelligence and data platform” founded in June of 2005 by Tristan Jehan and Brian Whitman
Tristan Jehan PhD from MIT has, in the past, done a thesis around the exact time EchoNest was founded. His thesis contained mostly what the “Trace-and-Hook” section of the project that John Seabrook is accredited to. From the thesis: “Machines have the power and potential to make expressive music on their own… solution models the life cycle of listening, composing, and performing, turning the machine into an active musician”
Colleague of Jehan, Brian Whitman, is also a graduate from the MIT Media Lab with a PhD in machine learning and music understanding. He was an artist/musician from the 90’s-2000’s.
We had the privilege of reaching out to Brian Whitman himself and discussing the heart of Daniels project, the 8 data points and how they came to be. Is this a transparent and open method EchoNest used?
EchoNest has it’s own analyzer that pieces signals from different audio files together. Points hat were easy to compute such as energy and loudness, were made from the analyzer. There was human raters in the process of rating the other data points, but it was mainly human raters just labeling audio files with what they thought fit the pattern, and the analyzer did the rest of the work. So, because multiple people worked on it, but, each had it’s own label thereafter being handled by the analyzer, the point rating system most likely has minimal margin of error between each rating and was rated on a fair scale.
Our opinion is, after talking to Dr. Whitman he explained how the eight data points came to be, the Methodology of this project by Daniels and Thompson, the methodology was clear and practices DHM values.
