I had my first stab at generating some sound today. It was a total flop: Not only does it not sound like the real thing, it doesn’t in fact sound like anything that is not random noise. (But I was pleasantly surprised to find that once I had all the data in .npy format, access becomes instantaneous so long as I specify 'r' in numpy.load(). Converting all those .vtu files to .npy was not in vain.) I need to figure out a way to make something that sounds less random. The question is how.

I finally figured out what’s wrong with my setup: I need to use the copy of python that comes with Paraview, which is a 64-bit version, and not the copy that came with MacOSX, which is a 32-bit version. I kept running out of memory not because the arrays were so huge, but simply because 32-bit python can’t handle these arrays. So I have converted everything to .npy files and I’m now ready to start try synthesizing some sounds!

I set my code to track 1000 random points, and after 2006 seconds (unattended, of course, and only at the 7th point attempted), it errored out complaining about not able to find the point it was looking for: The best point was over 2.0 pixels away. Let me try to relax the constraints a bit and see how it will do. But the question now is: How does Paraview do it?

So some sort of tracing code was in place and I ran it the first time with the full data set. The computer started crunching the numbers and then a few minutes later it spat out

python(54674) malloc: *** mmap(size=587386880) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug
Traceback (most recent call last):
…
MemoryError

I was only trying to trace three points and it still ran out of memory. I’ll need some other way to do this.

It looks like I’m now at the point where I’ll have to rewrite everything in python. I’m going to have to deal with multiple data files with multiple huge arrays, and perl is just not going to cut it. And then I’ll have to worry about interfacing with OSC. Of course, if I’m going to switch over to python, I should be able to use normal VTK functions and not worry about rolling my own file parsers. The problem is that two years after taking NLP at Coursera, I have already forgotten how to code in python. This is going to be a problem.