I decided to experiment with sound. While watching a movie, I realized that little sequences of music were used to indicate moods. I decided that synthesizing and recognizing sound would be something to test my new skills on. I downloaded a fourier spectrum analysis tool:
git clone https://github.com/vain/rtspeccy.git
The image is from a running trace of a sound. I thought that perhaps I could use finite element analysis to find the meaning in the medium. I thought that I could break the frequency and relative intensities into chunks in time, intensity and frequency as a 2D array that would be the items to select and combine for trees. Since there is no way to know what the right answer is, I decided it might be a tree within a tree and so I needed two speakers saying the same thing and solving for where their maximums intersect and how I change the data to make them resolve to the same structure. I did some Internet magic and found a set of 20 speakers, each saying a set of 10 different sentences and that should serve. Python has sound utilities and I can access the FFTW3 library just as well in Python as "C". It just makes it a little easier to script it in shell and Python and if it needs to be accelerated I can go to "C" and if absolutely necessary assembly, but the compilers are so good now it rarely benefits to do that.
It serves as a test for resolving some dimensional aspects of neutrinos and light or general EM from distant objects by automatic means. It allows me to apply the technique to identify patterns. It would even identify internal patterns to reveal a "voice" of a particular object like a particular supernova explosion category or many other relationships that become too complex to directly challenge with the amount of data available today. My hope is that I can pull "voices" out of FITS data bases of frequency, time and intensity data.
It seems that it would be sensible to solve a single temporal frame of elements as a factorial tree solution and then make each frame a factorial element in a second tree that grew sequentially in time, rather than attempting to do a factorial of all the parts at once.
I will also see if output from embrola and espeak festival can match the real voice data in some dimension.