They are making the files non-ascii for no useful reason
voice fast is related to speed.fast_settings. Others were errors in LoadVoice() parsing or commented out code. Note that some voices are from the original espeak and have comments that might be outdated.