Creating Better Compression for Audio Files

  NSF funds researchers who study developing a better music compression format
A team of researchers sponsored by the US National Science Foundation is currently investigating the way people and their instruments make music, in order to develop a new type of compression algorithm that would capture the very essence of a song.

A team of researchers sponsored by the US National Science Foundation is currently investigating the way people and their instruments make music, in order to develop a new type of compression algorithm that would capture the very essence of a song.

We are all familiar with the compressed files played by iPods and mp3 players. The only reason these types of files exist is because people want to carry as many songs as possible while occupying as little storage space as possible.

But compressing an audio file – therefore reducing it from hundreds of megabytes to only a 3 or 5 – means losing a large part of the audio quality it contains.

What researchers at the University of Rochester are trying to do is develop a method of doing this without having to renounce any of the “relevant” parts that make up a good song.

The team is made up of Mark Bocko and Dave Headlam, two experts at the university, who specialize in engineering, and music, respectively.

“We very quickly realized that the things he was interested in and the things I was interested in, in music theory, were actually very similar,” says Headlam.

The expert teaches music theory at Rochester Eastman School of Music. He has been collaborating with Bocko for more than a decade.

Bocko is a professor of electrical and computer engineering at the university, and also a member of the Rochester Music Research Lab (MRL).

The team is using NSF grant money to train a computer in recognizing how humans produce sounds through controlling their instruments.

“And so the whole idea [behind this study] is you want to capture the essence of the physics of how the instrument works,” Headlam says.

“So, what the computer learns is how hard they were blowing [test subjects using a clarinet], the blowing pressure at every instant in time, what their mouth clamping force was on the reed, and the fingering they used,” Bocko adds.

“But, it's really how the more subtle inputs and the changes of the blowing pressure over time, and how things are connected together. It is learning those parameters from a performance that is the essential part of this,” he continues.

Bocko says that a new audio compression format should be easy to produce, given that a human being could not possibly transmit all of the 1.5 million bits of information currently making up an mp3 file.

“I am a musician of sorts, and I'm interested in really understanding what makes good music good. And so it's been a way for me to indulge my interest in music, by working with my colleagues at the Eastman School and immersing myself in this,” the expert concludes.

Comments

By    16 Aug 2010, 11:05 GMT