1

I am trying to understand the working of an old computer speech synthesizer (SAM for the commodore64) which uses a series of rules to break English text into a string of phonemes. The rules make some frequent tests for membership of the following letter sets:

//  0x04        [ D J L N ]
//  0x08        [ B D G J L M N R V W Z ]
//  0x10        [ C G J S X Z R S T Z ]

Is anyone able to spot what traits the following sets may have and why they are significant?

Would I be correct in thinking the 0x10 set are fricatives?

Thanks in advance for any insight you can provide.

  • 2
    What are these? Letters typed in English? Codes for individual phonemes? How does the synthesizer work? What's the input like? – jlawler Feb 7 '17 at 16:42
  • Since it was said that the code is there breaks English text into phonemes and to sort letters into groups, I suppose these are actual letters rather than symbols encoding phonemes. – lemontree Feb 7 '17 at 17:25
  • For all letters, there are associated 'flags' which denote that the letter belongs to a specific set (the meaning of each set is what I aim to deduce.) One letter can belong to multiple sets. So above, the letters D,J,L and N all belong to one set, which the computer marks with the number 0x04. As an example, here is a rule defined by the system: ``` '.(S) =Z' ``` – Aidan Dodds Feb 7 '17 at 18:01
  • If the rule on the left hand side of the = is matched then the Phoneme encoded in text on the right hand will be output. As best I can see, the '.' here will match any of the characters in set 0x08. The ' ' character on the right hand side of the parenthesis will match against any letter at all. The rule as a whole will match any letters in set 0x08 if they are followed by a 'S' character and any other alpha character. I'm not sure if that will clear anything up, but hopefully is shows a little more of the context for this question. – Aidan Dodds Feb 7 '17 at 18:06
  • 2
    This resembles regexp methods of decomposing segments into features, without particular concern for phonetics. There must be more that just 3 such sets: the "meaning" of the sets would emerge from looking at what all of the sets are, e.g. what other sets is "D" a member of? "R" is in x'08, x'10, "L" is in x'04, x'08, so inferring the "meaning" would require having the total list. – user6726 Feb 7 '17 at 18:41
1

I don't get the full pattern, but there are some common properties:

  • D [d], L [l], N [n] are all voiced alveolar consonants; J (usually pronounced [dʒ]) is postalveolar
  • [ B D G J L M N R V W Z ] are all voiced; [ J L M N R V W Z ] are also sonorants, but [ B D G ] are not
  • C(when pronounced [s]), G (when pronounced [dʒ]), J ([dʒ]) S ([s] or [ʃ] in "sh"), X ([ks]) Z ([z]) are sibilants; however G when pronounced [g], C when pronounced [k] and R ([ɹ], or [ɻ],), T ([t]) don't quite fit in there and I don't know why S and Z occur twice in the row.
    They are not all frictavices since [t] is a plosive, [dʒ] is an affricacte, [ks] is a velar plosive + alveolar fricative, and [ɹ], [ɻ] are approximants
| improve this answer | |
  • This is a really nice insight and exactly what I was looking for, very interesting :) – Aidan Dodds Feb 7 '17 at 18:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.