Finally managed to apply myself well this afternoon and this evening to finish off the gaussian classification assignment for my Speech Recognition Masters subject. Basically, the assignment involved using R and Emu to analyse some speech data (860 short Australian English vowels from 10 different speakers) to train, optimise and evaluate a Gaussian language model using different feature sets. The various feature sets investigated for their utility in classificationwere vowel formants, cepstral coefficients and Bark scaled spectral bands.

Interestingly, although using the first and second vowel formant is a very simple approach, for the particular constrained problem I worked on, they provided the highest accuracy (about 82%). Note, however, that this was only for classifying short Australian English vowels (no diphthongs or long vowels, let alone consonants).

Anyway, it’s nice to have another assignment behind me. The reason I managed to finish it today is because I’m taking a few days off work, in lieu of the massive amount of overtime I put in to get SciFly finished. So hopefully I’ll make some more progress on my other uni work before the week is out!

Unfortunately, my 512Mb Imation USB 2.0 Swivel Flash Drive failed on Friday. It is actually the prize that I won for my best presentation award at the CSIRO ICT Centre Science and Engineering Conference in October, which makes it just that bit crappier having broken it.

I’m not sure what happened to it – it worked fine on Thursday night for backing up the SciFly data from our hired PC, but when I came to plug it in to my laptop on Friday morning, it simply isn’t recognised. I also tried it on my desktop PC with no luck either. I rang Imation technical support, and their “support” was decidedly underwhelming. Apparently I have to track down where it was bought from, and return it to the place of purchase, even though it’s still under the “1 year limited warranty” (what exactly is a limited warranty anyway?). I’ve started doing this, but it depends on the right people being able to find the receipts from October (given that it was bought as a prize).

In brighter news, after sending an email about my tale of woe to a few people (trying to track down the receipt from the original purchase) I got an email from our Communications Manager on Friday night in which he said:

Please go out and buy the finest, sexiest USB key you can find and buy it for yourself … I will re-imburse you.

Thanks for your incredible efforts at CeBIT and in the lead up

Excellent hey? It’s nice to have someone recognise the effort that I put in for CeBIT – it was a hell of a lot of work. It was of course very satisfying to have everything work out so well, but it’s also nice for people to make a point of saying thanks. Recognition of a job well done is easy to provide, but easy to forget. When people make the effort to say thanks (properly and sincerely), it goes a long way in my books.

So now I’m trying to work out whether to trust my data to another Imation or not. I really like the design of their product, and up until this mishap I’ve been very happy with the Swivel Flash drive. But now I have hesitations, both about the robustness/reliability of their products, but more so about their reluctant support. Does anyone have recommendations for a nice USB flash drive?

I didn’t have time to look at this for CeBIT, but it seems that there’s a simple way to manage the RMI server programmatically, rather than having to rely on external scripts to ensure that an RMI server is available on the appropriate machine.

To start the server on port 1099 (the default RMI port), just execute:

Registry reg = LocateRegistry.createRegistry(1099);


And to stop it again on shutdown, try:

UnicastRemoteObject.unexportObject(reg,true);


Simple eh? The only thing to be careful of is that the LocateRegistry.createRegistry() method returns the Registry object itself, not a stub (as the other RMI methods do). I’m just going to add this into SciFly, to make it one less thing for people to think about when running SciFly (to remind you, I use RMI to communicate between the barcode scanner, user interface servlet and natural language generation engine components).

We finally found time to test our new Dagger Trinidad sea-kayak on the Lane Cove River yesterday afternoon (pictures are not of us, but give you an idea of the scenery). That, of course, was after working out how to fit a 5.5 metre long boat on the roof racks of our Subaru, which is substantially shorter than that We need to buy some more of our very cool kayak straps to make things a bit simpler in the future (for attaching the kayak to the roof bars, tow hooks at the front, and tow-ball tongue at the rear).

Anyway, we found a great launching spot on the Lane Cove River, only a short distance from our house, thanks to the Lane Cove River Kayakers, who publish their meeting location on their website. It’s a perfect spot to launch from – lots of open space, plenty of parking, and as I mentioned, quite close to home. That should remove some of the inertia from getting out on the water regularly, as we’ll have to do if we have any chance of finishing the Hawkesbury River Classic!

We had a great paddle for an hour and half or so, paddling down-stream towards the harbour past Lane Cove, and then back upstream again. Our kayak is great! Very comfortable seats, feels very stable in the water, and has very easily adjustable foot pedals in the rear, and foot braces in the front. The only downside is that it’s a bit heavy to maneuver when it’s out of the water. I guess that’s the disadvantage of plastic boats. Still, lifting it on and off the roof racks will be good for our muscles

Tired. But very happy with how SciFly performed today, and with the interest and enthusiasm it has generated both amongst influential people in CSIRO, and amongst various people who used it today. I made some changes to the 2D barcode scanning code this morning, to make it more robust, and overcome some of the issues that we’d experienced in the first couple of days. Largely, I made the serial port reading code more robust, by making it wait for more data, making a bigger buffer, and giving a longer timeout to receive the barcode data. This, combined with more error tolerant processing of the incoming data, meant that we had absolutely no problems with badges today.

After showcasing SciFly to Alex Zelinsky yesterday, I also managed to get Trevor Bird (our Chief Scientist), Gary Doherty (our head business development manager) and Steve Giugni (our deputy director) use the system, and all were very enthusiastic about the possibilities for commercialisation. Sounds exciting!

I think we did a really good job of evangelising and explaining the technology to a very wide range of people. In fact, I think we’d make reasonable marketers if it came to the crunch!! Over the 3 days, we generated almost 300 tailored flyers, which I think is an excellent result. I also found out this morning that Ross Wilkinson, our Information Engineering Lab Director, is willing to create some funding to install a SciFly installation at CSIRO reception at Marsfield (and hopefully Macquarie Uni too, if I can be convincing enough!), so that would be a great step forward too . I think it’s fair to say that our hard work has paid off, and that we’ve managed to impress a very wide range of people with our work.

Tired – that’s the over-riding feeling from day 2. Another day with plenty of interest in SciFly, including from Alex Zelinsky, which is great! Looking forward to having some time off after all this is over though!

Had a quick (and early) dinner with Mum tonight at the Blackbird Cafe at Cockle Bay Wharf, which was nice. The drive home through peak hour again wasn’t as nice :-/

SciFly was very well received at CeBIT today, which was a bit of a buzz . We had the inevitable glitches now and again (some of the 2D barcodes on people’s CeBIT passes seem slightly non-standard in terms of their data format – or at least what we’re expecting), but nothing we couldn’t work around or quickly fix. Overall, we generated more than 100 flyers for interested people, which I think is a damned good effort (that’s more than 1 brochure every 5 minutes across the entire 8 hours of public CeBIT attendance today).

Had a small amount of time to wander the halls, but there was (as usual) not terribly much of great interest. A lot of consumer wireless products, a bunch of e-learning stuff and the usual pc cases with neon lights (though less than last year from what I saw). There were a few interesting applications – enterprise search engines, XML based content delivery stuff and the like. Another early start and long day tomorrow – we’ve got Helen Coonan, Minister for IT, Communications and the Arts [or something like that] opening the Future Parc exhibit at 9am, then a bunch of delegates visiting the CSIRO stand after Alex Zelinsky makes some opening remarks. Then, of course, another 9 hours of CeBIT. It’s fun talking with people and talking about our work, but oh so tiring!

Just got home from the energy-sapping task of setting up our CSIRO stand at CeBIT, in the Exhibition Centre at Darling Harbour. As per usual, it was a frustrating task, waiting for the stand being built around us, waiting for power and networking to be hooked up to the stand and so on.

Finally got SciFly setup by 6pm – it’s working with the printer and network connection, although the printer was a bit of hassle. It turns out that it’s very easy to crash HP ColorLaserJet 4600DN and 4650DN printers (and I mean hard lock, requiring a physical power cycle of the printer). These are big, enterprise model printers too (think something that’s about 1 metre tall). All you have to do is send it a valid PostScript level 2 file that has margins specified to bleed to the edge of an A4 page. Voila! Some internal postscript processing error, stack overflow and a hard-lock of the printer. So that took some hunting down to work out what was causing problems. The colour matching is also pretty crap, but it only makes the brochures look a brighter blue than they should, so I guess it’s not terrible.

As per usual, CSIRO hasn’t put any chairs on the stand! Argh!! So I can see another 3 days of tired legs coming up …

Drove home through peak hour tonight, and boy do I feel lucky not to have to do that on a regular basis :-/.

Finally managed to dissect the truly horrible Hawkesbury River Classic website and extract some information about the event. This involved reading raw javascript and HTML to reassemble the intended links – I hate crufty webpages that have been abandoned! The paddle sounds like a loooong way – 111km . Lots of training to do before October!

Still have to find time to show mum our kayak while she’s here visiting!

SciFly is a demonstration system for CSIRO’s Myriad
framework for information retrieval and delivery. SciFly’s first public
outing will be at CeBIT May 24-26.

SciFly delivers customised content based on user-selected interests.
In the current system, content is assembled into a dynamically generated
document and printed on-the-spot. Simultaneously a PDF document
is emailed to the user along with a plain text summary in the body
of an email.

This means that users receive relevant information presented in
contextually relevant formats:

• Paper flyer for immediate perusal
• PDF version for retention of information and later reference
• Plain text summary for mobile device access

The current demonstration of SciFly captures user details such
as name, email address and affiliation from a bar coded tag (as
supplied to conference delegates).The user then navigates a touch-screen
menu system to select topics of interest. The topics in this case
are based on the capabilities, projects and application domains
of the CSIRO ICT Centre.

Once all data is collected, SciFly assembles content tailored
to the specific set of interests expressed by the user, dynamically
fits the content to the space constraints of a double sided A4 page,
and adds relevant contact information, web links and higher level
context.

Importantly, SciFly does not just assemble pre-configured content,
but dynamically adjusts the amount and detail of content based on
the range of topics selected.