24 January 2025
Study on sonic dominanceProducers Set the Tone in Hip-Hop
Photo: Tim Ziemer
They used mel frequency cepstral coefficients (MFCCs), which are commonly used in music practice, to determine the sound characteristics of the songs more precisely. MFCCs are characteristics that describe the so-called spectral balance of a song– that is, how bassy, brilliant, or dull a song sounds.
The researchers also used a so-called goniometer. This is a typical analysis tool used in recording studios to determine the spatial dimension of a mix. “With the goniometer, we can find out, for example, whether the instruments sound dense like in a chamber orchestra or widely distributed like in Mahler’s Symphony of a Thousand,” explains Dr. Tim Ziemer from the Institute of Systematic Musicology and lead author of the study.
New approach to music analysis: The “sound map”
This data was then fed into an artificial neural network, which is used in machine learning, a branch of artificial intelligence. This made it possible to create a sound map from the similarities in the soundscape, with similar sounds being close together, and different sounds further apart. This made it possible to find out whether Dr. Dre, Rick Rubin, and Timbaland sound distinct or vary their style from song to song. The result was clear: “Each producer has their own typical sound—in terms of both spectral and spatial balance,” says Ziemer.
The producer’s sound profile also remains dominant in the interplay with the rappers’ voices. Songs that Nas produced with Dr. Dre or Timbaland do not sound typically Nas but reflect the style of the respective producer. Even a rapper’s distinctive voice seemingly does not influence the typical sound profile of a producer.
Science and developers could benefit
The aim of the research team led by Dr. Tim Ziemer was to use AI to make music analysis objective, causal, and comprehensible. Their innovative analytical tool is able to identify sonically dominant elements in the music production process. Their tool allows for further analyses—for example, in other genres—and could be used by streaming platforms to improve their recommendation algorithms.
“I assume that streaming platforms’ recommendations based on similar producers could be interesting for listeners and, possibly, even more relevant than those based on similar performers,” says Tim Ziemer, who—together with his team—has made the tool publicly available.