I can see some differences on the two allophones lists, Speakeasy and Amsoft Speech.
Almost all of them match, even the representation of the sound, of course there are sounds that are not represented on the two list. Speech has more sounds, but Speakeasy has more variations of the same sound.
Speakeasy chip sb0256:
http://courses.cit.cornell.edu/ee476/Speech/SPO256-AL2.pdfAmsoft Speech list:
"%" Pause 2
"1" to "9" stress or tilde on vowels and "/H". "9" lower note, "1" higher note, contrary to what it may seem.
You can put a number just after every vowel. I thing Speech can almost sing.
"." End of line
"?" Interrogation
"AI" "ER"
"A0" "EE"
"AH" "EH"
"AY"
"AW" "OO"
"AE" "OW"
"AA" "OY"
"OH"
"UU"
"UH" "/H"
"UW"
"UX"
------------------------------------
"D" "CT"
"DH" "CH"
"DR"
"DU" "Z"
"ZH"
"T"
"TH" "S"
"TR" "SH"
"N" "B"
"NX"
"R"
"L"
"M"
"V"
"K"
"P"
"W"
"J"
"Y"
"F"
"G"
Who has peeked the disassembly code could have realised that the sounds are formed of playing sample chunks of 63 bytes(but almost all of them measure 64) a number of cycles from 1 to 9.
There are complex sounds composed of two or three different samples.
But the most strange thing is that there are two chunks that are double than normal. The extra sample code is not used by the playing routines, so I think they are discarded sounds, not used on the commercial release.