MuseWire COLUMN: Imagine a world in which you receive money every time your music is played. There are firms that appear to be working to make this dream a reality. Representatives of Soundmouse, Landmark, ASCAP and APM spoke about the intriguing possibilities on a panel presented by the California Copyright Conference.

Titled “Digital Recognition Technologies: How Do They Work and How Are They Being Used?” the evening event focused on the software and machinery being used to identify music tracks by their digital signature.

As mentioned in prior columns, there are two basic methods of audio recognition: watermarking and fingerprinting. With watermarking, data is embedded into a song file. Although inaudible, the identifying data is accessible by computer. You might think of it as a silent and invisible bar code.

With fingerprinting, there is nothing added to the file. The song is scanned and analyzed, with the findings stored in computers. In his 2003 article, “A Quick Review of Audio Fingerprinting,” Wes Hatch wrote: “Simply put, audio fingerprinting is the process by which we are able to identify some piece of music, given just a few seconds of input to go on. The ability to do so is quite an astonishing feat, one that requires a complex level of math and the compilation of huge databases.”

Secrets
The companies offering digital audio fingerprinting are all using “proprietary methods,” which is a polite way of saying the companies don’t want to discuss how their systems work. Techie types with whom I spoke hinted at the use of Fourier coefficients, Hidden Markov Models, Mel-Frequency Cepstral Coefficients, evaluation of spectral flatness, linear predictive coding analysis and other terms I accused them of making up.

Whatever the process, the companies state they can prove that their systems work. Do they work a hundred percent of the time? No, but few systems are perfect, and the tracking of song files is so sorely needed that many music publishers, songwriters and performers would probably settle for a fifty percent success rate at this point.

The watermarked or fingerprinted information can also include the same type of meta-data text you normally place on an MP3 file: song title, composer(s), copyright, publisher, contact information, etc. Obviously, this can allow for more far-reaching and more accurate identification and reporting of the use of music in all media.

Panel of Experts
Many of the questions from panel moderator Steve Winogradsky, a CCC past president and current partner at Winogradsky/Sobel, probed the effectiveness and efficiency of the new technology. Joining Winogradsky and CCC President Shawn LeMone (ASCAP) on the dais were Mark Vermaat of Soundmouse, Darren Briggs of Landmark Digital, Lynne Lummel of ASCAP, and Adam Taylor of APM Music.

Soundmouse
Based in London, Soundmouse is a cue sheet management service utilizing music recognition technology to ABC, Discovery Communications, Fox, Disney ABC ESPN TV, ITV, BSkyB, and the international arms of NBC Universal and Sony Pictures Television, among others. As their website states, they perform “automated recognition of music in difficult audio environments such as music buried within television broadcasts.” Vermaat points out that Soundmouse systems “are designed to handle the large-scale reporting demands of the broadcast industry as well as to make certain audio recognition is accurate and complete, with processing scalable at an affordable cost.”

Landmark Digital Services
In 2005, Broadcast Music Inc. (BMI) purchased Landmark Digital Services. In so doing, they obtained rights to the patents of Shazam Entertainment Ltd. (Yes, the same Shazam many of you use in your cell phones to identify a piece of music you hear in a club, bar, restaurant, or wherever.) Part of this acquisition was the digital fingerprinting methodology called BlueArrow. As they describe it, “Using a patented algorithm, unique identifying features are mapped for all content.” Those mapped features are then compared to “program material gleaned from media broadcasts monitored around the clock.” Ultimately, “each fingerprint match delivers a title, identifying the exact time, date and source of the audio.”

ASCAP and MediaGuide
Mediaguide was founded in October 2002 by the American Society of Composers, Authors & Publishers (ASCAP) and ConneXus Corporation. Using fingerprinting and watermarking, they track music and advertising on 2,500+ radio stations in the top 150 U.S. markets. They state that their “audio and video fingerprinting technology can be embedded in broadcast head-ends, consumer devices and software applications. It can support asset tracking, community connections, copyright compliance, network quality assurance as well as creating new revenue streams through business intelligence, targeted interactive advertising and commerce.” Translation: you can make money from your music if you are in their system.

APM
Naturally, music firms with the greatest number of songs have a larger interest in automated tracking of their use. Since APM has a library that includes more than 300,000 titles, it made sense to have their president on the panel. Taylor, who is also on the board of the Production Music Association, has an interesting bio on his firm’s site: “For over two decades, Adam Taylor has been helping intellectual property companies, organizations and individuals manage and extract value from their copyrights, trademarks and patents.”

Who Goes With What
The three performing rights organizations each work with competing firms for music tracking. ASCAP is aligned with MediaGuide, BMI with Landmark Digital Services, and SESAC with Nielsen’s BDS and TuneSat. SESAC, in fact, has been active in this area prior to the other PROs.

Hunter Williams, a Senior VP at SESAC, provided written comments distributed at the event. “Today SESAC (through BDS) monitors over 1,600 radio stations,” notes Williams,” while pointing out that they are also in “a strategic partnership with TuneSat,” a fingerprint technology firm monitoring more than 100 TV channels. TuneSat originally planned to attend the CCC event but had to cancel due to scheduling conflicts. It is my hope they will post their views in the comments section below.

Pointed Questions
Winogradsky probed for the weak spots in the various methodologies. Vermaat pointed out that all sorts of music usage can have an impact on music recognition, including time stretching and audio processing such as EQ and compression, but also that audio is often of short duration and/or layered when used in motion pictures and TV programming. “The challenge for any fingerprinting technology in this field is to still make an accurate recognition regardless of these challenges,” Vermaat stated. He and others on the panel stated that “fingerprinting technologies in general have issues with these challenges.”

Lummel pointed to a hybrid approach to tracking, using technology to determine which songs were being played but often relying on cue sheets to determine precise usage, such as the length of the use and nature of the use, both of which are used by the PROs to calculate royalty earnings.

Taylor and Vermaat both noted that some of the new systems are helping generate more accurate cue sheets. “The cue sheet isn’t dead,” notes Taylor, “but the technology is a new way of getting data into the cue sheet.” He also pointed out that problems with inaccurate cue sheets might not be resolved until “months or even years after the production” is completed.

“After a decade of industry attempts to streamline the process, tracking performances on TV is still a largely manual endeavor,” writes Williams. “Cue sheets are notorious for being filed incorrectly, late, and in some cases, not at all, which adversely affects payments to songwriters and composers.”

It’s not just cue sheets that can be a problem. Improper input of track data can cause difficulties, something I know from personal experience: on the “Electro Bop” album by The G-Man is a song in the Golosio catalog entitled “Sheena Sez.” It was inaccurately entered by the record company as “Sheena Sex.” Digital audio recognition may help rectify these types of errors because the song can be fingerprinted and will be recognized no matter what title is in the meta-data.

Winogradsky brought up a number of points that could possibly give a false or incorrect reading by the technologies, including the re-titling of tracks (a common practice in the music library world), and the use of pre-recorded tracks, loops or samples. Briggs started to describe how Landmark uses a “heuristic recognition technology” that is overlaid atop their regular technology, but did not go very far with the explanation. After all, that would entail revealing proprietary information. But all parties recognized that the techniques mentioned above cause problems in accurate recognition of the correct piece of music.

Questions from the audience were strong and pointed, especially when one of the attendees referenced the competing systems being utilized by the three different PROs. Lummel, in particular, was pinned down by the question, “Would you accept data from Landmark?” The room fell silent. Finally, after a long pause she said, “Maybe.” Speaking for songwriters, composers and music publishers, I see this as a hopeful sign.

Future Uses
As Wes Hatch has written, digital audio data recognition might have been helpful in dealing with illegal Internet transmission of audio files, citing Napster as a prime example. “With an effective implementation,” Hatch notes, “administrators would be able to prohibit the transmission” of material protected by copyright. Again, we can hope.

Article is Copr. © 2010 by John Scott G – all rights reserved. This version is also Copr. © MuseWire, a publication of Neotrope®.