Scientists have managed to shrink a music file to one thousandth of its MP3 size

The music file contains a clarinet solo playing for 20 seconds and is encoded in less than XNUMX kilobyte. The invention was made possible thanks to two important innovations: the creation of a simulation that simulates a real clarinet on the computer and a simulation of the clarinet player himself

Researchers at the University of Rochester have managed to create digital music in a file a thousand times smaller than a standard MP3 file. The music file contains a clarinet solo playing for 20 seconds and is encoded in less than XNUMX kilobyte. The invention was made possible thanks to two important innovations: the creation of a simulation that simulates a real clarinet on the computer and a simulation of the clarinet player himself.

The researchers announced the new achievement at the International Conference on Signal Processing and Acoustics held in Las Vegas. It is not yet a complete reconstruction of the original clarinet performance, but the researchers say that they are getting close to it.

"This is actually the way humans create music," says Mark Bocco, professor of electrical and computer engineering and co-creator of the new technology. "There should be no need to measure the music thousands of times per second, as we do with CDs. People can't manipulate their tongue, fingers and breaths that fast."

When the computer plays the music according to the new file, it practically reproduces the original performance, based on everything it knows about clarinets and playing clarinets. Two Boko doctoral students, Xiao-Xiao Dong and Mark Sterling, measured and estimated the different ways in which the sound of the clarinet changes - starting with the pressure exerted on the barrel of the instrument when playing the various notes and ending with the way the sound propagates from the instrument. They then created a computer model of the clarinet. The result is a virtual musical instrument based entirely on acoustic measurements from the real world.

After the researchers created the virtual clarinet, they went on to create the virtual player. They simulated the way the clarinet player manipulates the instrument, including the position of the fingers, the force of the blow and the pressure exerted by the lips in order to determine how the virtual clarinet would respond to all of these. Now, Bocco says, the computer simply needs to 'listen' to a real clarinet performance to decode and record the various actions needed to create a certain sound. In order to reproduce the original sound, the computer feeds the list of actions it recorded to the computerized model of the clarinet player. The result is a very close reproduction, even if not perfect yet, of the original sound.

"We are still trying to integrate the way the tongue works, to understand, for example, how the player hits his tongue on the mouth of the clarinet in order to play notes in staccato [separately from each other, R.C.]," says Bocco, "but in music with long notes More that merge into each other, the method works well and it is difficult to distinguish between the computerized sound and the original."

The current method is only able to handle one musical instrument, but other researchers at the University of Rochester's Music Research Laboratory managed to find a method capable of distinguishing between the sounds of several instruments played together. A combination of the two methods may lead to a new way in which we can record and play music, with a minimal amount of saved information.

The researchers believe that when the method is perfected it may give computer technicians more intuitive ways to create music, by including the actions of virtual musicians in computerized synthesizers. Boko claims that in principle the method can be generalized for human voices as well. The models that simulated the human vocal cords and airways are indeed extremely complicated, but if he is right, in a few years we may hear new singers on the radio, who will be completely computerized. Any computer freak will be able to be proud of Pavarotti's rich vocal cords, with the right software to download (with a full license, of course) from the net. The most complicated operatic works will be transferred into the computer and will receive new touches. And all this, of course, with a minimal amount of memory kept on the computer. As Bocco says, "I think we've found the method that requires the least amount of information possible to play a piece of music."

And we can only wonder - will the day really come when we hear Baba-Luba and Om-Kol-Tom singing the duet from 'Phantom of the Opera' to each other? And is it really worth looking forward to such a day?

The answers, as usual, will only be found in the future.

Source (with the audio samples of the original piece of music and its recording using the new method) :