Hi, I am curious what other people are using for providing dynamic content on their stations. A few years ago a friend and myself setup a part15 station and I began writing some scripts to generate text-to-speech reports from various sources. At first it was just local gas prices from gasbuddy.com, then I included the weather, RSS feeds from AP and slashdot.org, local movie listings, local current events from websites in our town, and many other features. It uses random sentences in it's output so it's not as predictable and boring like you would expect a robot voice to be. It can also mix in background music into the reports. We had to take the station down over a year ago (lost access to the backyard we where running it from) so it hasn't been actual in use for a while but I have been continually working on it since that time. I'm about finished with a major re-write that should make it easier to maintain and I was curious if any other part15 operators would be interested. It requires a linux host and medium skill level to configure and operate (it requires several other software packages to do the text2speech and other functions).
The generated reports can be in wav or mp3 format. When the station was operating I posted freshly generated output files to a web directory and had the playout system (Windows box running Zara) download them about 5 minutes before playout using the Windows Scheduler and wget. Every hour we would broadcast the latest gas prices and the current weather. At 8am and noon we had news and other information.
There are two more large features I have had my eye on, but I likely won't implement until our own station is back up. I found someone else's perl scripts that allow you to tell the weather (rain/storms) over a specific area of land using up-to-the-minute USGS weather image data (currently weather is pulled from weather.com so it's very generalized over an area). While a little corny, the ability to provide accurate weather forecasting for a 1 square mile area was a fun idea and drove home the point that part15 listeners can be better served by your station ;). The final project on my mind for the station was a tie-in with an Asterix open-source PBX. This would give listeners the ability to dial a phone number, get a list of songs via a menu, and make a request which could be carried out by the playout system. Of course automated contests and other fun things could be arranged. But like I said, those two things are a little beyond my interest until our station is back up (And yes hooking a part15 station to a PBX is a little extreme - but if it's worth doing, it's worth over-doing.)
Anyway, if you are not afraid of unix/linux and are interested feel free to drop me a line and in a while here I'll have some scripts and directions available.
Thanks,
dosman at packetsniffers.org
1680AM, The Ocho!
Artisan Radio has experimented with automated speech - we used it to broadcast the status of the ferry to our audience (always of paramount importance to an island community). Unfortunately, the quality left a lot to be desired - the best application we found was Alive Text to Speech, with the Neospeech voice Kate. The ferry application worked, because the speech domain was restricted (we generated the text with a C program), and if something didn't sound quite right with the correct spelling, you could just use phonetics. I don't think the technology is quite ready yet for prime time if you have an unrestricted domain (i.e., a fairly large and uncontrolled set of phrases). That being said, we're still experimenting with using it to identify the last music played - whether we actually end up using the results is open to question.
Early on I had voice quality issues too, I found that using the latest version of Festival (1.96) along with an add-on package created by the HTS working group of Japan gave us quality comparable to the voice used by the US weather radio service. Both are open source and available for free. I have to admit that I even listen to the news podcasts generated on my lunch break, they are very good quality.
As far as the unrestricted domain, I've had surprisingly good results with what I call dictionary files using sed. It's one of those things that takes time to hone, but in a few sessions you can have a dictionary trained for about any source. I use a dictionary for street addresses and another one for generic sources. It won't ever be 100%, but 99% is easily achievable.
I put a demo of some of our reports along with some of my notes on setting up Festival here:
http://ibmgeek.shacknet.nu/ocho/demo