System and Method for Automatic Drum Track Generation

Technology #30992

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Collections
Image Gallery
An illustration of an example graphical user interface (GUI) allowing a user to select and evolve the CPPNs
Categories
Researchers
Kenneth Stanley, Ph.D.
Michael Rosario
Amy Hoover
Managed By
Andrea Adkins
Assistant Director 407.823.0138
Patent Protection

System and method for evolving music tracks

US Patent 7,964,783 B2

Software system and method for computer generation of drum rhythm patterns that are original and natural sounding for any song. The technology uses two primary functions. It first generates initial tracks from user-defined inputs. It then uses IEC, allowing users to both visually and aurally evaluate generated sounds, to refine and further evolve drum patterns until the user is satisfied with the pattern obtained. The technology can take an existing song without percussion as input and allows the user to evolve drum track accompaniment that automatically conforms to the underlying original tracks.

One of the major challenges of computer-generated music is sounding natural and inspired. Attempts to produce computer-generated drum tracks have been carried out using readymade drum patterns found in drum machines, keyboard hardware and commercialized software. An increasingly popular set of methods use interactive evolutionary computation (IEC). In these methods, users are allowed to input desired parameters such as timing and rhythm to influence the generated patterns. They then use a fitness function at each evolutionary step, to select a set of preferred, pleasing patterns. This method is time consuming, since the user has to evaluate many patterns including several trivial ones. In addition, since most computer-generated patterns focus on short musical phrases, rather than the whole song, looping is used to carry through to the end of the song. Consequently, computer-generated drum tracks are not song specific and the rhythms produced often sound artificial, uninspired and lacking in global structure.

Technical Details

To address these deficiencies, UCF scientists have developed an intelligent system and method, which produces natural-like drum tracks for a whole song. This system uses a combination of NeuroEvolution of Augmenting Topologies (NEAT) technologies, artificial neural networks, IEC, and a special representation for generating regular patterns called Compositional Pattern Producing Networks (CPPNs).

Benefits

  • Can be utilized by users with little knowledge of music or understanding of percussion
  • Is interactive, evolutionary and generates natural-sounding patterns
  • Can provide an infinite number of real-time elaboration capabilities, unlike static drum systems
  • Can generate drum track for specific songs

Applications

  • Individual composers and recording studios can use this technology to generate drum tracks. This system could reduce production costs, and facilitate experimentation with a variety of elaborations of the tracks, without having to employ percussionists. It can also be used to enhance music quality, by adding instruments and different musical parts like harmony and melody and can facilitate more exciting remixes of existing songs. In addition, the drum patterns generated can be exported to other music composition tools.


Additional Technology Numbers: 31231