Connect With Us

Enter your email address:

Delivered by FeedBurner




« Bob Marley: The Transformative Sound of Selling Out | Main | What If You Owned Music Think Tank? »

UPDATED: Can Computers Analyze Songs For Hit Potential? 

This an update to a similar post that appeared on Hypebot last week.  You may also want to read the post by Jay Frank that covers the same subject.

In response to the ASCAP/HITLAB announcement that basically endorses the use of algorithms to analyze the hit potential of songs, I thought I would weigh in on the subject.

Proceed with caution…
As someone that spent the better part of a year evaluating similar algorithms, technology, services, business models and patents connected to acoustic analysis and hit potential measurement, I can tell you that you should proceed with caution when making a purchase or career decision that involves the utilization of services that sell computer-based, hit-analysis technology. 

It’s fascinating technology, however…
Generally speaking, the technology is reasonably accurate (my experience: 80% accurate, and often close enough to my expectations) when it comes to plotting a song relative to a cluster of preexisting hits and then rolling the plots into a meaningful score.  However a high score doesn’t mean you have a hit on your hands, or that “hits” even matter anymore.  Read on…

Here are some pros and cons to consider when evaluating services that use computers and algorithms to evaluate music:

Computer-based hit analyzing technology - the pros…

Targeting.  If detailed reporting is offered, this technology should show you how close your song is to clusters of previously recorded hits.  This information is useful for targeting listeners of similar sounding hit songs. 

Selecting.  You should also be able to use the information provided to evaluate which of your songs has the most market potential.  Provided that you believe: historic success is a reasonable indicator of future potential. 

Filtering.  This technology is also useful as a filter.  Even if it only meets (average) expectations four out of every five tries (80% and then along a declining slope), in the absence of something better (“better” could be built), algorithms can definitely cut the size of the haystack down for someone looking for the needles; especially in a world that creates and uploads over 1,000,000 recordings a year.

Supplemental information.  For professionals analyzing songs, with the right reporting/presentation, computerized hit analysis is great (or at least interesting) supplemental information when paired with market /social traction data, crowd-sourced vetting data and detailed acoustic analysis/comparisons. 

Computer-based hit analyzing technology - the cons…

Songs that sound like they have been professionally produced or recorded only.  The last time I checked, hit predicting technology was not very useful for evaluating singer/songwriter demos. 

Just because it sounds like a hit…
There are numerous business and social factors that make a song a hit.  (Read the Song Adoption Formula on Music Think Tank.)  Business execution and promotion weigh heavily within the hit building formula (if there is such a thing).

Lyrics matter…  The technology I previously evaluated did not analyze lyrics, although lyrics as text or as acoustic features can be compared and analyzed by machines.  Make sure any service you buy can distill out the difference between lyrics about barking dogs, tuna fish and angry girlfriends.  Your epic song about cracked concrete may sound like a hit, but…

Connected to bullshit… This bullet is not a condemnation of the technology as much as it is a denunciation of the way I have seen this technology positioned and pitched to artists in the past.  When anyone sells you exposure based upon obtaining a “hit” score or anything else, go to (comparatively speaking, Compete is accurate enough) and verify the exposure potential of the, site, label or service.

[This is the update to my previous post on Hypebot:  Upon further consideration, I do believe it’s possible that someone could offer an artist some form of exposure based upon a score.  However, before you ever sign away an inch of your rights, make sure you consult an entertainment attorney, interview artists offered the same deal, and stop to think about the scale and durability of the exposure that is being offered.  Be mindful that most exposure is fleeting: here today and gone tomorrow.]

Old paradigm thinking…  Do hits really matter?  When it comes to songs, determining popularity potential (along a spectrum and within niches) and then matching songs to taste preferences, and artists to target audiences (through recommendation), are the technological advancements that should really matter to the majority of artists (IMHO).

The bottom line: you can learn something by using/applying “hit” measurement technology wisely,  Just don’t use it for all the wrong reasons.

about Bruce Warila

Reader Comments (8)

A similar analogue version to songwriting was devised by Claus Zundel, a producer most famous for crafting Sydney Youngblood's early 90s hit, If Only I Could.

I seem to remember him asserting that he had taken elements from the most successful recent hits and assembled them to make the song. Sadly, for Claus, he didn't manage to cannibalize his own success successfully again, to the same extent.

Any computerized analysis would have to take into account so many variables that it makes it all but useless.

(What I would be interested in would be some software that provides personal info on radio producers' private lives that I could use to blackmail them into playing one of my releases on heavy rotation until it's composition, performance and production become totally irrelevant.)

One use for it, though, would be a way to convince labels that they are signing up sure-fire successful artist, so as an elaborate con it might work well. Often, if a company take a keen financial interest in any artist they can put a train of events in motion, powered by money and old-fashioned record-working, that ends up with a hit.

Interesting, though.

December 12 | Registered CommenterTim London

For the same reason computer analyzation can't get translations right, I defy any machine to have the heart, soul, cleverness and subtlety to analyze lyrics ... This kinds of programs are such a sham.

December 12 | Unregistered CommenterTonsoTunez

It's entirely possible to count up the frequency of occurrences of words and phrases for every song that ever charted and then plot a new song's lyrics against that 'graph'. Done repeatedly and over time, meaningful correlations are most likely going to appear. Outliers will probably fall through the cracks (?), but the information presentation will probably be useful (or at least interesting) to someone. Read more about this research:

December 13 | Registered CommenterBruce Warila

Thanks for a new word, there, Bruce.

I get the feeling if you remove genuine outliers (or if they fall through cracks) the info will be rather thin, to say the least.

The huge difference in production (from room to mic to compressor to, literally, atmosphere) alone would be enough to prevent recreation of any ideal circumstances that would lead to a track being enjoyed by most of its listeners, let alone making any money for its creators.

Like the CIA's experiments with LSD, though, there could be unforeseen results that would make all this hard work worthwhile.

I can see a whole raft of tunes that, like the Top Of The Pops series of covers in the 70s, could become a whole genre all its own.

December 13 | Registered CommenterTim London

Tim, you know about Pont-Saint-Esprit?

Bruce, excellent article, I really appreciated the summary of where the field is at. I think the comparison to translation algorithms is pretty apt, but I also think (hope?) that Universal Translation is not far over the horizon.

Discussions about technology are in a funny, fucked up place in 2010 because tech journalism is all hype, tech users are just consumers, so it's easy to be cynical -- but the big trend is: the Borg always wins. These small, crude incremental steps will abruptly start scaling up and suddenly, we're living in another damn sci-fi novel, again. Deep Blue lost a hell of a lot of matches on the road to defeating Kasparov.

It's easy to see how within 10 years this technology could evolve from "metrics without context" into some legitimate hit-making, hit-identifying technology. Compare the web metrics we were getting in 2000 from our hosting companies to what's available on google analytics today: data that is immediately useful for actual businesses. That's definitely where this kind of software is headed, too -- it will be a long road paved with failed businesses and hilarious press releases.

December 15 | Unregistered CommenterJustin Boland


I have been working on this post about diminishing subjectivity... It begins:

I feel like a number. I'm not a number. I'm not a number. Dammit I'm a man. I said I'm a man
- Bob Seger (1981)

December 15 | Registered CommenterBruce Warila

I needed some universe translating to get the references - OK, google, wiki and an ever-trusting disposition. Good story, true or not.

December 15 | Registered CommenterTim London

Thanks for this blog very much. I am a big friend of this blog. I think this blog is liking by everyone. Thanks for a new word, there, Bruce. Thanks for it.

PostPost a New Comment

Enter your information below to add a new comment.

My response is on my own website »
Author Email (optional):
Author URL (optional):
Some HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>