Document how much memory espeak needs, & possibly influence

Espeak will not synthesize speech into the wav file if a large amount of memory is not available when pressing the Speak control button. A simple illustration of this is to boot an OLPC XO fresh, use FreeIconToSpeech, then launch a few other applications, then try to use FreeIconToSpeech again. If the other applications used significant memory, then espeak will not synthesize speech, returning a generic error code of -1. This error code currently is output to the command line, and visible when you run ./free_icon-to-speech.py from a terminal window.
The top command says python is only using <40%, but this is not a very accurate picture of the memory usage. Even when running top's refresh rate many times per second, espeak does not show a significant memory spike there.

You can also reproduce the problem depending on how many concepts you try to load. This makes testing easier, so you don't have to reboot for each test. You can simply remove some of the categories or concepts in the xml files to decrease the total quantity, decreasing the memory needs of FreeIconToSpeech.

Perhaps once this is better understood, we could propose espeak improvements for it's use on smaller hardware platforms.