Here in early March 1998, we now have the first launch of a system designed to provide customized audio programming over the Internet. The company is Imagine Radio Inc.

Imagine Radio has several of the characteristics described in A Modest Proposal: Audio Programming From Online Music Catalogs, though falls far short of the ultimate vision presented there. Nevertheless it is an exciting beginning, just as Firefly's collaborative filtering is a promising suggestion of what might eventually become possible in the realm of content recommendation tools.

The key innovations of Imagine Radio lie in the personalization of the music selection process. Users may choose from 16 channels of music content, with around 80 selections per channel (these channels are programmed by human experts according to a genre-based model). The company provides a custom player based on RealPlayer 5.0 streaming audio format, to capture a series of selections within the channel according to a quasi-random algorithm. This algorithm can be further influenced by a user's ratings of a currently-playing selection, which changes the propensity for that track to be replayed in the future. Background information on the artists, etc. are also available for the tracks played.

Given this setup, we have the first genuine step toward actual customized programming of audio on an individual basis, and Music Unbound applauds the motivation behind it. Nevertheless, there are still some serious drawbacks connected to this system that prevent it from being the complete accomplishment of such a mission.


Randomized selection is constrained to channels that are defined by human judgment.

Though it is true that within the 16 channels there is a range of personalization that is offered, the definition of those channels is not accessible, but rather is directed by experts whose concepts of these genres is not open to user input. This is perhaps the most serious drawback, since it severely limits the actual range of personalization allowed the user, given that these genres are not absolutes etched in stone, but rather cultural artifacts defined by traditions and accidents of fate rather than by truly content-based considerations.

These experts are all chosen from the existing music industry, and thus are inevitably likely to follow existing industry programming habits. Although in concept the personalization capabilities of Imagine Radio are exciting, the real-life manifestation of these expert choices may largely work against that impulse, giving users hardly more than an illusion of customization.

If instead a variety of selection sets were offered on a more custom basis, such as via collaborative filtering or other methods, this would have a much wider range of personalization.

However, one might well forgive the company for limiting these selection sets in this way for now; the server power required to expand beyond this is quite extensive, and probably not feasible given today's available technology. This setup is designed so that over 1200 different selections might be streaming all at once, which is a hefty river of bandwidth in today's world. As a proof of concept, this is probably the limit to what one might reasonably attempt. It may fall far short of the several hundred thousand albums worth of content available from a catalog, but a first step is a first step. There is a long way to go, but you can't take a second step before you take the first one.


Integration with purchasing capabilities are not yet in place

Word is that the company is exploring partnerships with music retailers to cover that last piece of the puzzle, though perhaps first in the guise of links into music catalog sites, rather than integration into the audio service itself. One step at a time.

The business model for this service is based purely on advertising revenue, both from audio ads interspersed occasionally between the music selections and from visual banners that will appear on the player itself, onscreen. These ads can be targeted according to user preferences, and thus may eventually be coordinated well enough to user interests to become viewed as useful information (or at least not aggravating noise or interruptions).

However, the early stages of the age of subscriptions may well be upon us, and that extra revenue stream may well be able to eventually provide a much broader range of selectable material and methods to select it.

Though this first version is constrained by today's available platforms (including servers, network bandwidth, users' machines and selection choices), it shows a glimmer of hope that true custom selection and custom-casting of music audio might have a real future in the online world.

-- Dan Krimm, 3/98