Bach to the future: computer will finish composition
The mystery of how Johann Sebastian Bach intended to finish one of his last works may be solved not by a musical genius but by a computer program designed to spot musical patterns.
Computer scientists believe they have developed a system that can second-guess how the composer would have completed his unfinished fugue, a piece he wrote in the last decade of his life and may have tried to wrap up on his deathbed.
The incomplete work, published after Bach’s death in 1750, has long attracted musicians eager to finish what the celebrated composer began.
Scientists at the University of Washington in Seattle have created MusicNet, a system that enables computers to read large numbers of classical pieces and identify trends.
It works by taking synthesised pieces of classical music, which have imprecise timings but contain data labels that computers can understand, and mapping them on to real performances of chamber music. MusicNet alters the timing of the synthesised piece to match the real one, in a process known as dynamic time warping.
The combined piece is then entered into a library from which computers can ‘‘listen’’ to a piece and determine which notes are being played, by which instruments, and how the notes fit into the metrical structure of the work.
John Thickstun, a doctoral student working on the project, said the labelling had to be precise for computers to generate meaningful results.
‘‘You need to be able to say from 3 seconds and 50 milliseconds, to 78 milliseconds, this instrument is playing an A. But that’s impractical or impossible for even an expert musician to track with that degree of accuracy.’’
Thickstun said computers could use the data to finish Bach’s piece in the same way that they used image libraries to fill in the gaps in damaged pictures.
He said completing the fugue was one of his team’s ‘‘mediumterm’’ ambitions, after solving more modest problems.
‘‘The intent is for the computer to complete the fugue on its own without human intervention.
‘‘The interesting question for us is to teach a computer to do this in a convincing way.’’
Several musicians have attempted to complete the work, but there is no definitive version.
Sham Kakade, associate professor of computer science at the university, said that he hoped that MusicNet would spur creativity among composers and developers of artificial intelligence. ‘‘At a high level, we’re interested in what makes music appealing to the ears, how we can better understand composition.’’