Technology‎ > ‎

Algorithm charts the way to music success

posted 7 Feb 2012, 03:22 by Mpelembe   [ updated 7 Feb 2012, 03:25 ]

A team of scientists from Bristol University in the UK will be watching with particular interest as Sunday's Grammy Awards ceremony hands out statuettes for musical achievement in 2011. The scientists have developed algorithmic software they say can predict whether a song will be a hit or a flop with 60 percent accuracy, a development that could one day make the job of the Grammy judges much easier.

XL RECORDINGS - You might assume that Adele's international smash hit "Rolling in the Deep", is the result not only of her unique sound but also marketing and image. But a Bristol University scientist says the likelihood of a record's success can be reduced to a 23-character equation, developed in a machine learning algorithm devised by himself and his students.

Dr Tijl De Bie, senior lecturer in artificial intelligence, led a team who programmed every song in the official UK 'Top 40' singles chart over the past 50 years into a computer. They evaluated musical features such as tempo, time signature, and song duration.

They also computed more detailed summaries of the songs such as harmonic simplicity, how simple the chord sequence is, and non-harmonicity - how noisy the song is.

The team then developed the 23-character 'hit potential' equation that scores a song according to its audio features. The Dutch academic said songs

could be classified as either a hit or a flop based on this score, with an accuracy rate of 60 per cent as to whether a song will make it into the top five of the UK chart or remain below position 30.

Reuters asked De Bie to analyse three smash hits from 2011 - Adele's 'Rolling in the Deep', Bruno Mars's 'The Way You Are', and surprise UK number one 'Wherever You Are' by the Military Wives Choir.

"The features that contributed to a good score for Adele were its energy, so it was not too energetic, for the time when it was released. There was some variability in the beats and it was somewhat danceable, although not very danceable," he explained. The song topped the charts in a whopping 14 countries.

De Bie explained how the software analyses songs. He said: "The score of the song is computed as a linear combination of several feature values that are automatically extracted from the audio of the music. These feature values they relate to the very aspects, like the duration of the song, how fast it is, which is quantified by the number of beats per minute or the time signature automatically from the audio itself. There are more complex things such as the danceability of the song, how much energy there is in the song and the harmonic complexity which is a musical characterisation of how complex the audio is."

De Bie added: "We devised a machine learning technique, an algorithm that can be implemented on a computer which is able to tune these weights, these parameters, automatically by looking at past successes and past flops, so to say, on the charts in the UK."

The team say their equation incorporates subtle changes in the prevailing styles of particular areas. For example, before the 1980s, the danceability of a song wasn't that relevant to its hit potential, whereas today it is crucial.

De Bie explained that one of the biggest selling hits of the past quarter century, Kylie Minogue's 1988 single 'I Should Be So Lucky' would not be successful if it had been released today. The song moved at 120 beats per minute, fast by current standards, but made number one in seven countries, spending five weeks at the top of the UK charts, where it sold 650,000 copies.

The track was co-written by Mike Stock, one of the most successful songwriters of all time. Best known for being a member of the songwriting and production team Stock Aitken and Waterman that achieved success in the 1980s and 1990s, he has written and produced 18 number one records in the US and Britain, and more than a hundred Top 40 hits.

Stock poured scorn on De Bie's software. He said: "I would rather take issue with how you get your database together on that because, you know, you're going to be dealing with analysing thousands of hit songs and hope that the software's going to tell you which way to go next. If you're just doing something that the computer tells you to do because it's been a hit before you're not interested in that. You want to know what the next hit's going to be."

Although admitting that he liked the idea as "a bit of fun", Stock insisted he wouldn't be using the software himself.

"Why would you want to think that a software device has come up with something you're supposed to feel and believe in? It's music, it's supposed to touch you," he said.

The team found its algorithm to be more accurate for some eras than others. It was particularly difficult to predict hits around 1980, with the equation performing best for the first half of the nineties and since the year 2000.

When asked to rate Bruno Mars's hit 'The Way You Are' the algorithm's results were less clear than with Adele. "On all the features that we analysed it was never extreme in either way, so that means that the score that we got from it was quite an average score as well, although slightly to the hit side. So that means that we would have categorised it as a hit as well, but only very marginally so." In fact the track topped the charts in nine countries, becoming the ninth most downloaded song of all time in the UK.

The Military Wives Choir registered firmly in the flop category, but became the UK Christmas number one, although De Bie argued that the charity record, designed to raise money for families of British servicemen, scaled the charts due to sentimental reasons rather than any artistic or algorithmic merits.

He said: "That didn't do well on danceability at all, it didn't do well on loudness, it had a very clean harmonic sound, and harmonically, musically speaking, it was a relatively easy song, whereas most of the time the hits are slightly more complex than the non-hits. So you could say it's a surprise that it still became a hit but of course there is the social aspect surrounding it that in this case made all the difference. Probably it didn't become a hit because people liked the audio very much but because they liked buying it for a different reason."

De Bie says the work has revealed the steady change in musical tastes that have been reflected in chart music over the past half century. Analysis showed that music has become easier to dance to and louder over the years.

The Bristol team are in discussions with a number of companies wishing to commercialise the software. An iPhone app is already in development. Until then music fans keen to test the chart possibilities of their favourites songs can use the website

Whether the songwriters of tomorrow use it to gauge their song's hit potential or it becomes a short-lived entertaining teenage fad, time will tell. With the music business's premier awards ceremony, the Grammys, being held in Los Angeles on February 12, it's probably doubtful that 23-character algorithms will be among the factors taken into account by the team of judges.