Software used in music production software could be programmed to automatically detect music defects, researchers at Google have said.
In a paper published in the journal Science, Google researchers said they could use deep learning techniques to detect musical flaws, as well as to identify and fix them.
The work is important for companies trying to automate the production of their music.
For example, music-production software could help companies produce more professionally-produced music.
The researchers said that using AI to automatically identify music defects was already an effective tool for the detection of music flaws, but there was no good way to tell if an automated software was actually detecting the flaws.
They therefore used deep learning to develop a program that would automatically detect musical defects, as part of an experiment.
This is the first time that a research group has demonstrated that a new technique could be used to detect a flaw in an automated music production system, the researchers said.
The researchers are now working on improving the program to be more robust.
The research team, led by Professor Chris Beecham, of the Google Brain group, said they were not sure how long it would take the automated system to detect the flaws, and that they would like to test their software with live performances.
“However, given the limited time available, we expect to be able to do some work with it at some point in the future,” they wrote.
They also said the researchers were exploring other approaches to automatically detecting musical defects in music.
For example, the team is developing software that could identify defects that occur when the machine is unable to interpret music accurately, or when there is a poor understanding of how a song is played.
The team said that these defects could be corrected using manual training.
“We believe that these techniques are already useful in music training software, and we are exploring potential applications for these techniques in music detection software,” they said.
“However this work does not show that automated detection of musical defects can be performed automatically.
We would like the team to improve this system to demonstrate this.”
In the paper, the research team described their research as a way to find a better way of automatically detecting flaws in software.
“In music production, there are a number of types of flaws that could potentially lead to musical problems.
We want to know whether the system can detect and fix these faults, so that music can be produced in a more effective and efficient way,” they concluded.
The researchers have shown that their system could detect defects in an automatic music production program in a way that would be more reliable than manual training methods, and which would be simpler to implement.
The paper does not mention whether other researchers had used deep-learning techniques to automatically spot flaws in their own music production systems.
The new technique, called a “recurrent neural network”, was developed by Beechams team in collaboration with the team from Google Brain.
The research team has also published several papers in the last year about music detection and machine translation.