We assume humans have a finite set of distinct vowel and consonant forms.
So, take their spectra from a representative sample, the standard sonic spectra used in signal processing. Then using the miracle of a major transformation of which I do not remember, covert those to probability spectra, finite bandwidth. Then measure first,second,third and fourth moments of the distribution. Use these for a four dimensional space and encode the spectrum into a spread about the origin, weighted by frequency of use. Huffman encode by moments then pack a sphere. Each phonetic element ends up with likely a nine bit code, 1,2,3,3 bits per dimension is my guess. It becomes a four color problem, bu likely can be reduced to a three color.
Add a few bit, no problem, but you end up with a damn good phonetic alphabet, and as long as it is for bot to bot comm, then humans do not even care. Add a bit of grammar, rules for conversion to paper and rules for conversion to sound. But sound is no good without mouth, have rule to convert to mouth shape on the video along with vocalization and the effect is much better.
No comments:
Post a Comment