Print Story Retroactive Radio Recorder
Diary
By tmoertel (Fri Feb 20, 2004 at 08:08:39 AM EST) (all tags)
As I described in my online journal, I have created a simple "VCR" for the radio. It works surprisingly well, especially considering how little time I have invested in it.

It just goes to show how effective open source software is for building bespoke solutions. Linux + ALSA + Speex + cron + a small shell script = Radio VCR.

But that's just the tip of the iceberg. I have an idea for an



One big problem with any VCR concept is having to program the VCR with the desired shows. If you forget to program in your favorite radio show, the VCR won't record it, and you're out of luck. If a presidential address pushes your show back by seventy minutes, you'll record the president and not your show. If a special 90-minute episode of your show is broadcast one week, you'll probably fail to record the last 30 minutes.

And that's where the

< Live is good. Whodathunkit? | BBC White season: 'Rivers of Blood' >
How many concurrent channels? by jacob (3.00 / 0) #1 Fri Feb 20, 2004 at 08:46:34 AM EST
It seems like that could be a deal-breaker, or at least a complication. If you want to record the entire broadcast spectrum, seems like you'd consume HD space much more quickly than your estimate. Assuming you limit it to two or three interesting channels, though, seems very workable.

Another idea: once you've got this massive database, why not apply some voice-recognition to it to, for instance, extract all shows that mention "George Bush" or something like that? I can't imagine that it would be easy, but generic voice recognition has been a hot topic for a while and if you don't mind low hit rates you might be able to get something cool working.

Another: rather than (or in addition to) saving data locally, mix in a peer-to-peer system so that if the VCR doesn't have a radio program you want, it can ask around until it finds somebody who does.

Anyway, sounds like a cool project. Let us know how it goes.

--

Voice recognition/data extraction by spcmanspiff (3.00 / 0) #2 Fri Feb 20, 2004 at 09:16:44 AM EST
I've been tinkering, mostly in my head, with the idea of community-driven data extraction of this nature --

e.g., have a list of names that you'd like recognized, and some sort of trainable sound-byte-recognizer.

Then, a user can search the Big Sound Database for "George Bush"; but also can quickly check and reject false positives. That rejection goes back into the training data for the recognizer...

Misses are a bit more effort, but putting hordes of people to work on indicating name references that were missed should be do-able.

I had been thinking of something like this for doing NLP, using trainable software to recognize proper names in articles, then associate direct quotes and/or paraphrases (harder!) with them.

Then all I'd need is a DVD or two with the complete Lexis-Nexus news archive .... *evil genius cackle*

Anyway, the idea is that with enough user participation you end up with a huge corpus of 'known good' people saying "George Bush" clips -- as well as 'known bad' examples -- which you can use to (hopefully) develop a very very accurate "George Bush" recognizer.

[ Parent ]
Doesn't solve the delayed programmes problem by Gully Foyle (3.00 / 0) #3 Fri Feb 20, 2004 at 10:21:00 AM EST
You'd still have trouble jumping to a programme that ran late.

TV channels used to broadcast programme id information, so you could set your VCR to record a particular programme, rather than at a specific time (you tell it the time, and it waits for the correct id to arrive before it starts recording). Dunno if it still does. The info in DAB radio would let you do this I guess. Dunno about analogue radio.

Yes, it does, but I didn't describe it by tmoertel (3.00 / 0) #4 Fri Feb 20, 2004 at 10:27:37 AM EST
Basically, in the same way you can adjust a particular show's length, you can change also change its starting time. So if a news flash pushes this Wednesday's airing of your favorite show back by 25 minutes, you can correspondingly push back the corresponding recording's start time, and URRR will re-record that one show for you.

--
Write Perl code? Check out LectroTest. Write markup-dense XML? Check out PXSL.

[ Parent ]
But how would you know? by Gully Foyle (3.00 / 0) #5 Fri Feb 20, 2004 at 10:30:37 AM EST
You'd be flipping through last weeks Radio Times going hmm that looks interesting, and fetch it from your database, to find that it got delayed. You'd have to listen to quite a lot of broadcast to find how long it got delayed for.

[ Parent ]
It's fairly straightforward by hand by tmoertel (3.00 / 0) #6 Fri Feb 20, 2004 at 07:03:52 PM EST
When you discover that a particular airing of a program has been delayed, edit the corresponding recording's duration, making it, say, four hours long, more than enough to cover any reasonable delay. Then play it, skipping forward until you find the program's start. Look at the elapsed time, and that's the delay. Finally, return to the recording's properties, shift its start time forward by the delay, and return the duration to the program's normal running time. Now you've "repaired" the recording.

--
Write Perl code? Check out LectroTest. Write markup-dense XML? Check out PXSL.

[ Parent ]