How about this for an AI project: a computer system that takes the audio stream of a National Public Radio station and presents a commercial-free audio stream as an output, presumably with at least a one-minute delay.
It should be pretty easy for a classical music station. There isn’t too much talking and the non-commercial stuff is typically right before or after a piece. Even a program that simply muted humans talking and let through music would perhaps be sufficient (though watch out for Recitative!).
For an NPR talk/news station it would presumably be tougher because the same announcers present both content and commercials. However, that’s what makes it a great project for a Master’s student in AI if not a PhD.
What do readers think? Should a prize be established for the best programs that can do this? It seems like a useful application of AI and also something where it is easy to test and score programs (sum of minutes of content improperly muted and minutes of commercials improperly not muted, for example). Programmers around the world could compete for honors/money.
Since most commercials play repeatedly are a mostly fixed length, and NPR shows each have fairly fixed clock: http://www.slate.com/blogs/the_eye/2013/09/12/_99_invisible_on_npr_s_broadcast_clock_the_diagram_that_rules_public_radio.html
It should be a small matter of training?
If there was a database of commercials like the database of songs used to track copyright infringements, it would work. Most podcast announcers outside of NPR have moved to reading commercials instead of playing recordings, so it would take an algorithm which could determine meaning from language.
Better yet, get the FCC to stop NPR stations from abusing the non-commercial frequencies.
What if you owned a device that would allow you to carry a large library of classical music (any music, really) with you in compressed digital form, or to stream it from the cloud over cellular radio, so you would never have to listen to NPR fund raisers again? What if there were services that would serve up these streams to you ad free for a nominal subscription charge? And what if there was some technology, some sort of short range radio system, whereby you could send this music from the device in your pocket to the car radio speakers? Wouldn’t that be wonderful?
Great idea, Jack, if you want an unpaid job as a DJ. See this post about Apple Music for why it (and all similar services) don’t work for classical music listeners (unless what you want is to hear randomly mixed movements from different pieces).
I worked at a major public radio station for many years. You may not love it but “underwriting” helps pay the bills. Its rarely more than 90 seconds per hour (not sure if that’s an FCC mandate) and the messages must meet certain standards: no calls to action or claims about specific products. Compared to commercial radio it really isn’t that bad.
As someone pointed out, NPR news and talk shows work on a standard-length segments, so its pretty easy to see where the underwriting and other promotional blocks fit in, but music programs often have more flexibility.
Pledge (fundraising) drives are another matter. Only around 10% of the audience pays membership at most public radio stations and, IMHO, 3 or so weeks of disruptive fundraising per year is an inane, outdated model. Especially as more and more supporters switch to “sustaining” (continuing monthly) membership.
Some stations – notably, KQED in San Francisco – have experimented with interruption-free internet streams for paying members. Its a start.
Someone needs to come up with a better model. Until then, expect pledge drives and underwriting at NPR stations.
Why not do something like recaptcha with it? Ask a human or bot on a webpage to describe the content and use that to not only screen the audio stream but also weed out bots on the websites? There is your minute delay, I am sure it will take that long and possibly a bit more to get enough votes to categorize the content.
I think the comments don’t reflect the original posting. The assumption is that NPR doesn’t change and the unchanged NPR provides a great challenge for AI programmers.
How about crowdsourcing the input of when a commercial break begins/ends (getting people listening to the broadcast using an app to volunteer for free or paying them via something like Amazon Mechanical Turk) and broadcasting that signal in near real-time to users of the app (or automatically muting if listening to the broadcast through the app)?
I like Geraldo’s “distributed mute button” idea – the DMB. Patent it quick, before IBM gets to it: http://www.pcmag.com/article2/0,2817,2493155,00.asp?kc=PCRSS03079TX1K0000584