Record executives
have long relied on their golden ears, gut reactions and luck
to pick the hits.
But some music producers are trying to take the guesswork
out of this quasi-mystical process by resorting to
supercomputers, artificial intelligence and algorithms.
Hit Song Science, a high-tech system that electronically
compares newly recorded songs with a database of thousands of
old Top 40 tunes to identify common attributes, is being used
by Sony, BMG, EMI and other music giants to help them predict
which pop sounds will ultimately soar.
"Songs have hundreds of distinct mathematical
similarities," said Tracie Reed, vice president of North
American operations for Polyphonic HMI, which has developed
the novel note-crunching system. "We analyze those underlying
patterns, including melody, tempo, harmony and pitch, to
determine a song's hit potential."
The use of science to unlock the secrets of smash hits
comes as the recording industry and radio stations are being
slammed by some critics for promoting what they say is a glut
of derivative, homogenous music containing little artistic
impulse.
So will 50 Cent's next chart-topping hip-hop tune sound
suspiciously similar to an old Randy Travis country ballad?
Several songs that were put through Polyphonic's elaborate
testing have since become ensconced on Billboard magazine's
Hot 100, Reed said.
She declined to name the hit recordings because of
confidentiality agreements the company has with labels, some
of which are apparently sensitive to allegations that they are
manipulating the public's taste for profit.
However, two record industry executives said they lean
heavily on Polyphonic's pre-release reports to weed out
probable sonic duds.
The information "helps reduce uncertainty" about a tune's
prospects, said Muff Winwood, a Sony UK vice president and
singer Steve Winwood's older brother.
Producer Peter Swartling, whose projects and artists have
won him more than 20 Grammy awards, said he uses Polyphonic's
services in the pre-production process to keep him and studio
musicians on the right track.
"We as producers have an ability to hear what sounds right
and what doesn't," Swartling said. "However, I think we can
always become more precise."
Certainly, producers and songwriters have long tried to
decipher the formula for cutting best-selling records,
occasionally creating tunes bordering on plagiarism.
As early as the 1950s, Elvis Presley and Pat Boone were
accused by some listeners of ripping off Little Richard and
other black singers.
Likewise, George Harrison was successfully sued for
"subconscious plagiarism" over his huge hit "My Sweet Lord"
because it has the same melody as The Chiffons' 1963 classic
"He's So Fine."
Science fiction
But music industry observers say Polyphonic's initiative
takes the practice of breaking down songs note by note into
the realm of science fiction, where computers become the
primary creative force.
"This has an ominous sound to it, and I can understand some
people having a negative reaction," said Paul Williams, the
legendary editor of the rock magazine Crawdaddy and the author
of "Back to the Miracle Factory," a book analyzing 1990s-era
rock 'n' roll.
Yet while Polyphonic's technology has the potential for
creating music with a "boring sameness," Williams said, it
also could be "a stimulus for some appealing and enduring
recordings. . . . Making music is a mysterious process which
often comes up with works of art, so I wouldn't rule out any
peculiar circumstance involved in that."
Polyphonic, which has a U.S. office in Harleysville, Pa.,
maintains a supercomputer containing more than 250,000 CDs, as
well as a subset of about 700 of the top 30 songs from
Billboard's Hot 100 during the past five years.
The company's proprietary software purportedly isolates
dozens of sonic patterns contained in past hits - including
brightness, tempo and chord progression - which likely
influenced music consumers' purchases.
Hits are also evaluated on the basis of total sales,
highest chart position and date of release.
"These songs form distinct (statistical) clusters" in the
data, Reed said, noting that a Beethoven symphony could have
at least some of the same hit components found in a U2 song.
"New releases, potential releases and even unsigned acts can
then be compared with the database to allow a record label to
see how well songs fit into the current market."
Polyphonic issues each new song a final grade based on a
numerical scale of 1 to 10. "A rating of over 7 means the song
has a very strong hit potential," Reed said.
Labels are paying Polyphonic $3,000 a pop to evaluate
potential releases.
Intangible factors
The company said its high-tech system isn't completely
foolproof because it has no control over such intangible
factors as a record company's ability to promote a song or the
inherent charisma - and sales power - certain recording
artists possess to push a song, any song, to the top.
Moreover, Polyphonic's computer does not assess a song's
lyrical content.
"Besides our rating, a record still needs a professional's
ear and promotion," said Reed, noting that the company's
rating system has occasionally given a song hit-bound status
even though it was clearly a long shot.
One such tune was a six-minute instrumental, which was "a
really good song, but was inappropriate," she said. "We told
(the client) to shorten the song, add lyrics, and it might be
a hit."
Polyphonic said it plans to soon provide its service online
to amateur artists and garage bands that need a quick read on
their prospects for the big time.
Artists will be able to get a Hit Song Science rating for
an as-yet-undetermined price.
The company is also talking to music retailers about
installing computer kiosks so customers can quickly compile
playlists containing hundreds of songs reflecting their
personal, quirky preferences.
"They may indicate that they like 'Start Me Up' by the
Rolling Stones and that they want five songs similar to it,"
Reed said.
Reed said Polyphonic has no intention of using its musical
X-ray machines to write and produce music, as some cynics have
suggested.
"While we help show producers and songwriters what's
working and not working in their compositions and production
work, this technology does not do the work for them," she
said.
Frank Green: (619) 293-1233; frank.green@uniontrib.com
