Yeah... The Empeg isn't quite ready for primetime as an easily-extensible computing platform. Mlord's kernel was a huge step. The alternate root partition is another biggie. It will let us keep user programs intact between software upgrades, and as I understand it, will provide the user space portion of the seamless launch procedure we're looking for where you just add programs to config.ini. Mlord has done most of what needs to be done in terms of allowing for text from config.ini to be available to the user apps. I'm sure there's other work to be done there.

If people are looking for constructive things to do, I think that the following efforts can go on in parallel with these efforts. Some of these suggestions might already be in the works, if so, please let us know how you're doing!

- an efficient graphics API
possibly written in ARM assembly
- a "storage API"
take smallish chunks of data and save it to / load it from raw sectors on an unused partition such as hda2 or hda6
- a sound API
take PCM data and mix it in with the player output (it appears Kim has a good start on this)
maybe even a routine to mix in an MP3 file (a nice bell/whistle feature)
- a nice installer package for user programs
along the lines of Frank's but one which circumvents Emplode and transfers the package directly

I had this one running with ARM linux on my ipaq:
http://www.speech.cs.cmu.edu/flite/

bobo

Well... Here's the deal. There are basically two options for TTS on Linux. Rsynth and Festival. Rsynth is much lower quality and doesn't do as good of a job figuring out sentence structure or the proper sound for a given letter. Festival is what is used in the demo MP3. Festival unfortunately has a memory footprint of 10 megabytes. Even if we had the RAM I don't think we'd have the power to do real time text-to-speech. I did cross-compile Festival and the binary runs, butt immediately crashes on my Empeg.

Rsynth would be an option except that I can't get it to compile and run properly on my Empeg. I need to hack it up to write exactly 4608 bytes and I haven't had any luck with it. Anyone smarter than me who wants to give it a try is more than welcome to do so. If someone can get rsynth working then we have at least rudimentary TTS.

If it used Rsynth then I'd have to do a lot more manual fiddling with the questions (replacing some E's with the "schwa" phoneme, for instance) because rsynth isn't as intelligent about choosing the proper sound. Me parsing through 20,000 questions is not likely.

The authors of Festival have decreasing the memory footprint listed as one of their goals, but I'm not holding my breath.

Ahhhh... I could NOT find that one in any of my searches. I was even looking at CMU's speech site! Thanks for the URL. Downloading now...

20,000 questions * 10k = 200 megs. Not sure anyone would want to download that much.

That's only 1/3 of one percent of the capacity of a 60 GB empeg. I think there are a lot of people who would give up that much space! Or, another way of looking at it, it's about three CDs worth of space.

tanstaafl.

Okay this is another victim of the 4608 byte requirement on /dev/audio. Running Flite gives me this:

empeg:/usr/local/flite# ./flite "hello, world"

on-4608 byte write (256)

failed to write 128 samples

So I'm not great at all this low level IO stuff, but I found the #define for the audio buffer size was set to 128. So if 128 samples = 256 bytes (16-bit sound) I figured increasing the #define to 2304 (4608 / 2) would do the trick. Naaah. For some reason it only writes 1584 samples at a time, and I get this:

non-4608 byte write (3168)

I'm not smart enough to figure out the rest. So flite has the ability to output to a WAV file. I did that and copied the WAV file over to my PC and the quality of the voice synthesis definitely leaves little to be desired compared to the full version of Festival. I guess that's what makes it "Festival Lite." For instance in the example I posted, it says the verb form of "contest" instead of the noun form (accent is on the second syllable when it should be on the first.) The real Festival handles it without problems.

So if we want anything good sounding, we need to use sound files rather than realtime synthesis. At least for now...

Sure I realize that. I've got a 40GB Empeg so it's no big deal. But downloading that 200 megs is a little onerous... I dunno. That example question isn't necessarily the longest question in the database either, I'm not sure what the mean is, but the average question could end up being 15 or 20k instead of 10k. Then if you wanted it to say the answers aloud, that'd be even more.

Then there's the fact that I don't know how to play an MP3 file from my software, much less mix it in with the Empeg's output...

But it's something to think about. A neat feature, certainly not what I'm going to focus on, though.

I have some mp3s that are 200 megs.

Calvin

Yah. 200 megs isn't that bad, which is why I am considering the pre-recorded soundfile route for this after seeing that realtime text-to-speech directly on the Empeg might be somewhat far off. I am going to experiment some more with Flite and at least try to get it correctly outputting sound to the audio device before I totally give up, though.

Im trying to install emptriv and having problems. I have uploaded the file to my empeg and extracted it to /drive1/emptriv
It seems to have all extracted ok with a questions directory, but when i execute emptriv it doesnt do anything. I have tried leaving it for a few minutes and breaking out of it, but then nothing appears in the hijack menu.

Any help appreciated.

Are you running it in the background? You can't break out of it and still have it bound to the menu (the process ends!)

You need to type emptriv& to run it as a background task. Then fire up the player (exit back to the shell-player loop will also work.)

It should then appear in your menu.