e.g.
Link 1 – Intro/Sponsor Credit
Track from your library
Track from your library
Link 2 – Weather for your area today
Track from your library
Link 3 – Topical Link/News
Track from your library
Link 4 – Throw forward to ‘live’ station content

The smart playlist would contain variables so you’d hear your most played or highest rated tracks (or on the flipside, different music each day) inbetween topical content relevant to you and tailored to the duration of your commute. Ideally you could choose from different presenters or channels and commercially the links could be bumpered with sponsor credits (as opposed to ads which could be skipped).

I tried it out with edited portions of links from a show (Richard Bacon on Xfm IIRC) and it worked quite well but I just don’t have the programming skills to pull it off unfortunately.

I guess booting up the pc each morning, downloading and docking my iPod is a habit I’d have to get into and it might have limited appeal but somehow, I think it might work!

Cheers,

Bren.

We’re building the project you’ve described. Most of it is up and running now, at RadioWeave (http://radioweave.com/). I hope anyone interested in building a personalized, modularized, multi-source radio will test it out and give some feedback on whether we’re getting it right.

For a couple of years we’ve been trying to tackle the issues you bring up: how to create a radio program containing only the bits any particular individual is interested in, for the mutual benefit of the listener and the content creator (and the union of those sets).

Here’s a youtube video of me driving to work (as if I didn’t actually just work at home) listening to my unique RadioWeave variety show and describing each bit of audio as the radio delivers it. http://www.youtube.com/watch?v=IhQhPHJCIuE That’s my experience on one particular day. Everyone’s show will be different.

Our paradigm for how to personalize the radio experience is for each user to subscribe to multiple micro-channels, where each micro-channel represents a particular interest. For example, car-racing, turtles, news about Sri Lanka, local weather, local traffic, hip-hop music, and so on. A user can also create microchannels composed of music they own, recordings of friends directly into the system, audio-bloggers (e.g. from audioboo, cinch, soundbiter…), all tracks with a particular tag, blogs converted via text-to-speech (through odiogo or hearablog), tweets and facebook messages robotically read. Really, we want to get anything available as audio so that the “radio” experience is just as personalized, relevant, and hyperactive as any desktop web experience.

The part of the process for bringing all of these elements together in a way that is interesting to each listener, that process you describe as “a bit of magic”, has turned out to be a lot more complicated than we originally expected. To facilitate content creators to make appropriate “modcasts” we are encouraging more tagging and shorter podcasts, and chapterization of longer podcasts, but until we get more buy-in from content-creators we are using some automatic sound processing and breaking up of long podcasts so any particular segment is generally no longer than 6 minutes. Another important part of working with content creators is to make sure they get their metrics about listeners, location, feedback (through thumbs-up/down/skipping). Of further interest to content-creators is the ability to interact more directly with the audience through voting or direct replies (aka “backchannels”) from their computer or device microphone (so that you can participate, even in a time-shifted way).

The bulk of our content currently is straight from podcasts. If a podcast isn’t in our system already then any user can add that podcast for the benefit of all. I hope those is only a temporary solution, and that your idea of “modcasts” catches on; then our piecing together of multiple content types will work even better for the users and the broadcasters.

Our premiere platforms for listening are the web site and the iPhone application. We’ve also released a chumby basic player and have android and blackberry versions nearly complete. We have recently published the server API, and much of our client code, so that others can participate in creating more clients (or making better ones that ours), and creating tools for adding more content. (Anyone who wants to work on this area with us, let me know.)

Where the current RadioWeave differs from your description is that we’re concentrating our limited resources on a live experience, and not the one that would be downloaded to an offline such as an iPod. This is important for getting real traffic reports or news breaks, for instance, or hearing twitter buzz within minutes of your friends tweeting it. We do have a tool in the labs for creating the off-line iPod experience, via a standalone apple script that interacts with iTunes (so that it can correctly get heard, skipped, and votes), but that tools is still looking very inelegant so we haven’t officially released it.

I’ve gone on too long, sorry, but it’s quite exciting to come across someone describing so well a shared vision. The above is just a brief introduction to where we are and where we’re headed with RadioWeave. Any feedback is greatly appreciated.

This is really interesting – at Sheffield we developed technology that could analyse speech transcripts and use this analysis to produce speech recordings of any length by progressively removing irrelevant content. Whilst this analysis was purely statistical, there’s no reason that some element of personalisation could be included. So high modularisation may not be necessary – a podcast player could automatically determine which portions of the recording were of interest and just deliver those.

Another thing we explored was using high level “gists” as a means of leading people to content they were interested in. So you could take an hour long news broadcast, play a gist and allow listeners to drop out of the gist and hear the full content – all of this being an automatic process. I think large vocabulary speech recognition is at a point where these technologies will soon be relatively simple to implement (we got good results on meetings data) and so it would be interesting to see where this kind of stuff goes in the next few years….

Andrew