Dagstuhl Seminar 24302 Theme Song
A medium-tempo funky song about music informatics researchers meeting in Germany with Irish-style melodic lines played by bass, drums, acoustic guitar, piano, vocoder, baritone sax…
This text prompt, originally hand-written on a whiteboard, was the starting point for a music experiment conducted at the Dagstuhl Seminar 24302: Learning with Music Signals. With AI-generated music emerging as an important theme in our talks and discussions, some participants came up with the idea of trying out state-of-the-art deep learning music generation for a specific purpose: creating a theme song for the seminar.
Title, lyrics, and music of the resulting song “Funk and Data in Deutschland” have all been generated by an AI with limited human intervention (see details at the bottom of this page). You can find both the AI-generated audio and the official debut live performance below.
Funk and Data in Deutschland
Created by Bob Sturm using Suno and Udio.
Back to the Real World
Directed by Christof Weiß, the song was performed live by 13 seminar participants at the traditional seminar concert. In preparation of this performance, a lead sheet with lyrics and chords, as well as a horns arrangement were derived from the audio recording.
- Lead sheet – lyrics and chords (PDF) by Bob Sturm and Christof Weiß
- Horns arrangement (PDF) by Christof Weiß and Christian Dittmar
Debut Live Performance
Recorded at Schloss Dagstuhl in Wadern, Germany on July 25, 2024.
The Process
The process began on Suno using the prompt above. Suno generated the song title, “Funk and Data in Deutschland,” along with lyrics and two initial audio files.
| Audio Example: Suno #1 | |
| Audio Example: Suno #2 |
The lyrics were nice, but Bob wasn't too excited by the music. Neither one fit the intended instrumentation. So he engaged Udio, where the prompt was slightly modified to: “A medium-tempo funky song about music information retrieval researchers meeting in Germany, featuring Irish-style melodic lines played by bass, drums, acoustic guitar, piano, vocoder, baritone sax, accordion, melodica, and other percussion.” From this, Udio produced new lyrics and audio files.
| Audio Example: Udio #1 | |
| Audio Example: Udio #2 |
Using the first verse of the lyrics generated by Suno instead, two more Udio songs were generated.
| Audio Example: Udio #3 | |
| Audio Example: Udio #4 |
Bob chose Udio #4 as the starting point to extend the song step-by-step, synthesizing 32-second extensions with the lyrics that were originally generated by Suno for each verse, chorus and bridge. This resulted in the following output.
| Audio Example: Udio #5 |
Finally, he added an intro and outro using the “automatic” lyrics mode in Udio.
| Audio Example: Udio #6 |
The final track was lightly edited in a DAW by shortening the intro (removing 0:21 to 0:34 from Udio #6), as well as applying some compression and equalization. The entire process, from the initial prompt input to the final song, took approximately 40 minutes.