Audio Programming
From Online Music Catalogs

The light may already be peeking out at the end of the tunnel. It requires true broadband custom-casting on the Net, but that may not be too far away. (Yeah, yeah, they said that about cable modems a couple years ago, and they're still piddling around with the regulations and technology standards, but once we get the secure transaction standards for e-commerce, perhaps the gate will open.) Even if it's further on down the road, eventually it'll happen.

And then, this is what will happen. I hope. First, some background analysis.


What exactly does music radio provide?

There are two services that music radio provides to its listeners, in the guise of an audio programming service:

1. A fairly consistent variety of music for life's auditory wallpaper

2. A free sample of newly released records prior to purchase

The record business mainly makes use of the second aspect. The radio business itself concentrates more on the first, as a means of attracting a large audience with which it keeps its ad rates high, to make a profit from ad sales. But I'm not really that concerned with the radio business itself. I'm more concerned with an audio programming service in the abstract, from the user's point of view, and how this might dovetail with the desires of record companies and independent acts to sell their products.

Note: There used to be a third angle, which still lives on in non-commercial radio programming, i.e., education/information about music either in general or more often by genre, in little corners of programming by college, public and off-shore "pirate" stations. But the audiences for these programs are currently not large enough to make a significant impact on the music business by affecting mass airplay charts. They manifest too much variety to register on the radar for any one artist or composition, with little if any tie-in to measurable sales. Successful commercial programming long ago migrated to "formats" that are specifically non-exploratory, using an increasingly narrow selection of tracks for high-rotation airplay that creates quick brand recognition for individual songs and artists, leading to clear short-term "hits" in the business.


What does a listener want from audio programming?

A listener wants two seemingly contradictory things from audio programming: consistency (they want to know that when they tune in a particular channel, they will get "that kind of music" that they expect), and freshness (within the confines of "that kind" they would like to hear some new things that they haven't heard before, even while they like hearing some of their old favorites along the way as well).

The standard way to approach this in a mass-market medium is to define genres (or even more narrowly in today's world of radio formats) that determine the bounds of what to include and what not to include, in order to reach a desired consumer target market that is attractive to consumer product/service advertisers. Then, the average of the majority rules: anything that has too much "personality" to avoid repelling significant chunks of the target market is excluded. Can't put in a single cut that makes people switch channels. Music that is undistinguished enough to be widely tolerated will suffice, while any music with character that is potent enough to evoke a significant amount of passion will inevitably turn off enough listeners to affect ratings.

So this means that the least common denominator of a target demographic audience is all that can survive in this mass medium. That ultimately doesn't serve the individual listener (we know that each person has their own cross-section of music that they like and dislike, that generally diverges quite a bit from what even those with similar demographics would choose for themselves).


What does a record company or act want from audio programming?

Pretty simple, they want to get heard by the people who would like their music and want to buy it. The averaged-out formats in music radio constitute a tiny bottleneck through which newly-released music can get heard, and leads to a tiny cadre of successful acts in the midst of a sea of back-listings in a record company's catalog that never even cover their production costs.

Record companies would like it if all of their acts could recoup the production costs (and promotion, and tour-support, etc.), because it would wipe out the losing side of the business and focus entirely on the profit-making aspects. Even if not all the acts recoup, the more sales that come from unrecouped artists, the higher the unit revenue, since the record company is keeping all the royalties (artists don't start to get royalties until they are recouped for all advances on production/promotion/etc.).

Bottom line, it's in the company's financial interest to have as wide a base of artists from which it gathers revenue. Instead of 10-20%, think what kind of money a record company could make if it had 70-80% recoupment. However, radio formats work directly against this goal.


How could audio programming be improved?

Customize it to the user on an individual basis, without regard necessarily to well-defined genres (certainly not formats). Then listeners get to hear more effectively what they really want, and the population of listeners gets a chance to address a wider variety of artists, in the total mix of their individual tastes.

It's like each listener could define their own "custom channel" on the spur of the moment, unconstrained by anyone else's arbitrary definition of music genres or styles. The listener wins, and the record company gets a more accurate reach for the true market of each individual act in their domain, allowing more acts with moderate audiences to rise to their full market potentials.

Instead of today's all-or-nothing, blockbuster-or-bust dichotomy, we get a truer continuum of economic levels, allowing the opening up of the "musical middle class" in the commercial music world, where a great deal of quality material is currently lost.


Okay, so who is in a position to build this?

Online music catalogs. Large stocks of hundreds of thousands of titles are now available from several online catalogs, many of which already allow you to listen (via RealAudio or other methods) to samples from their catalog. The first online/downloaded delivery of music for consumers is already here. Recommendation features, from collaborative filtering to experts' and artists' favorites to even algorithmic systems for direct, nonsubjective content comparisons, are beginning to become more populous. With broadband platforms of the not-too-distant future, this can become a personal, reliable, high-fidelity medium for customized audio programming.


What does it look like?

Here's the sketch. Record companies and independent artists cooperate with a music catalog to provide digital files in a proper format for loading into the music-sample database. (That is, if you want to be included, as an individual act, it is in your interest to help the catalog company avoid processing hundreds of thousands of albums on their own dimes. Spread out that cost; in the end the investment will be worth it.) The catalog company would provide a free software tool to correctly pre-format the data given to them. Load up a huge database of digitally stored music, that can be accessed multiply and on demand by the playback technology provided to the user, via the Net.

A user can log in and define a "custom channel" by selecting any track stored in the database as a "center" point. Then using any combination of recommendation tools, define a set of tracks or artists that are "similar to" the center point track. If you like, go out a second step, to find anything similar to any of the first "similarity set" -- that determines a pretty large set of material for a single listening session.

Now, program a "random walk" through this musical space. That's your custom audio program.

Some fine-tuning:

1. Provide a "skip" button, so that any track that comes on may be skipped if the user chooses. (Also provide a "pause" button to accomodate interruptions without losing content.)

2. Provide a "mark" button, to identify for future reference, if the listener really likes the track, and/or to bring up detailed information on the artists, etc. Allow seamless, integrated purchase of the track or album right then, in context. (Delivery either online or by mail -- subscriber info is already available.) Perhaps provide a discount on the audio service for frequent-purchasers, or for other user-feedback.

3. To fine-tune the musical space, allow particular artists or albums to be excluded from a selection set, and time frame to be specified (new releases, nostalgia, etc.).

4. Instead of running this all via desktop PC, create a custom appliance that hooks into this system like a regular radio receiver (cf. the cell-phone or cable TV model where the phone or set-top box is free and the service is paid). Build a decent transistor radio into it, if you like, with an AM/FM/Net switch (some of those "boutique" stations still program some pretty cool stuff, and we don't want to deny the listener the full range of programming options). Provide line-out jacks to hook this into a high-fi system, small speakers for remote listening (how about a wireless Net connection?), and a headphone jack. Plug and play. Give it away for free in return for a minimum subscription to the service (three months?).

5. Make sure your database is as large as possible -- the best service will be the one with the most choices for selection into a custom channel. If you are a single major label, operate this service as a separate entity so you can distribute the products from all the other labels (you can still get a small profit from those sales, and you get total volume staying with your own service instead of someone else's).

6. Advertising could be included, to reduce costs, particularly for music-oriented promotions such as concert appearances. Any music-related advertising could be captured in the recommendation features, so that they would be targeted to people who are on average more likely to be interested in such information. However, ads could be set up so that they cannot be skipped, as a guarantee that they are delivered whole or not at all (each ad delivered would reduce the user's cost by a fixed amount per time duration). Provide an option for the user to determine the frequency of ads in their programming (every third track, every fifth, etc.), allowing them to decide for themselves just how much ad exposure they are willing to endure in return for cost reduction. (And if well-targeted, some users may consider these ads to be useful information, and choose a high frequency, even as high as one ad per track. The key here is user empowerment to decide for themselves.)

7. For all parameters entailing user control, take care to determine a good set of default settings as a starting point, so that setup is not a huge task. Then provide prominent and clear instructions on how to increasingly take more control over the content.


What about pricing?

This needs to be balanced carefully. BMI/ASCAP/etc performance royalties would need to be negotiated for this unprecedented service (personalized, individually delivered, but not stored unless permanent product is purchased), in such a way that the subscription fee pays for it along the way (artists are paid for any significant portion of their music that is not skipped during the program). Ads need to be priced per delivery to compensate properly for this cost, plus leave a margin for profit by the service (price for advertiser = user's service discount + service provider's profit). Subscription fee needs to be flat, per month, to be able to realistically compete with free, ad-supported programming (with optional discount offers such as ads and volume purchases), low enough to be valuable for the user (no usage charges), high enough to break even plus margin.


What about security and piracy?

Use encryption and watermarking such as is built into systems offered by companies like Liquid Audio.


Doesn't this require tons of bandwidth for the server and pipeline?

Yes it does. But there are several strategies that can help reduce redundancy in the bitstream. Most of these are already in use now, and can likely be further improved and optimized.

1. Mirror sites -- the simplest way to scale server/pipe bandwidth is to create multiple copies, and spread out the users and/or content among them.

2. Non-lossy audio data compression -- given constant bandwidth, there are several ways to use less of it, without losing necessary information. One recently-conceived method, for example, uses transmission of "difference data" off of an initial "carrier waveform" rather than direct transmission of actual audio data. By choosing well-matched carriers for differing sections of music, the data bandwidth can be substatially reduced while hopefully still retaining fidelity.

3. Buffering -- at a certain point, bandwidth becomes large enough to exceed the minimum for real-time streaming for individual users. At that point you can begin sending "future data" ahead of time, stored by the user's equipment until it is needed, and allow for traffic variation over the Net, thereby stabilizing traffic somewhat (traffic peaks are one significant obstacle to custom-casting).

4. "Pseudo-custom-casting" -- combining the "random walk" and moderate-length durations of single tracks with Internet multicasting can enable increased efficiency with a "pseudo-on-demand" service, after usage rises to the point where there is significant duplication of content among simultaneous users. By setting up a large array of very fast audio-streaming servers that can continually loop a significant subset of the database of music tracks, one can simulate a real-time service. With randomization of start times of the variously-long individual tracks, chances are great that within any typical selection set there will be a new track re-looping every several seconds (adequate to serve as "breathing space" between tracks, as on radio programming).

Thus the "randomization" can arise simply by choosing the next available unplayed track in the selection set, at the end of each currently-playing track, and if several users have included that track in their sets and request it simultaneously, it is multicasted out to all of them together, thus conserving the bit stream and server bandwidth redundancy.


These are the kinds of things necessary for "broadband custom-casting" (aside from brute-force performance increases in individual servers and network connections) -- that was not an idle requirement. But if Moore's Law holds steady over the next several years while such efficiency techniques are developed further, such "audio on demand" could very well become feasible.

In the meantime, LAN/Intranet-based development with a limited user set can serve as proof-of-concept for those wishing to begin immediately.


What about the proliferation of "boutique" programming via the Internet?

This is beginning to happen, as many audio programmers are taking to the Net to avoid the bottleneck of airwave licenses, etc. (The installed base of receivers for this is still relatively small, as is the net itself, but undoubtedly this is on the rise.) The concept described here is simply the same idea of narrowcasting taken to its logical extreme, and given to the user to control, with a seamless tie-in to fan information, communication and sales, in context.

In fact, the record industry seems to be reporting a slump in recent years. However with the booming economy of the same period, one wonders if independents might already be eating into the total volume of recorded music sales, at the grass roots level, and thus unmeasured by the big players in the industry.

Custom-casting may cut into the volume of celebrity profits in the music business, but it probably will not wipe them out completely or even largely. In fact, format-oriented radio probably will endure indefinitely, though it may adapt itself to changing market forces created by a custom-casting market. All the above idea would do is open up the middle levels of audience size to artists who properly fit there, turning the hourglass shape of today's market into the more ideal triangle that it should naturally be.

-- Dan Krimm, 8/97

Copyright © 1997 Daniel Krimm. All rights reserved. Hyperlinks to this page are encouraged, but republication must be cleared with the author. Please observe proper citation etiquette.

---