Index of /knowledgemedia/SPEECH/AMIGA/TLPATCH
Name Last modified Size Description
![[DIR]](/icons/back.gif)
Parent Directory 11-Jun-2003 14:43 -
![[TXT]](/icons/text.gif)
EXCEPT.TBL 11-Jun-2003 14:43 11k
RENAME.BAT 11-Jun-2003 14:43 1k
![[DIR]](/icons/folder.gif)
SOURCE/ 11-Jun-2003 14:43 -
SPEECHTY 11-Jun-2003 14:43 15k
SPEECHTY.INF 11-Jun-2003 14:43 1k
TLPATCH 11-Jun-2003 14:43 12k
TRANSLAT 11-Jun-2003 14:43 7k
***************************
** TLPatch V1.0 **
*** Richard Sheppard - 1991 ***
** Toronto, Canada **
***************************
A utility to allow corrections in pronunciation for programs
that use the Translate() function. It is offered as "freeware"
without support (or liability) in return for my years of pleasure
with the Amiga. I hope it can be useful.
One of the unique features of the Amiga is its ability to
translate english text into spoken audio. This is a two-stage
process involving the Translate() function which converts the
text into phonemes, and the Narrator device which uses the
phonemes to synthesize the voice.
Unfortunately, this feature appears to be underutilized, due
perhaps to the many words which do not translate accurately.
This program allows you to extract the exception table from the
translator.library, and using a text editor, edit the table
then restore it back into the library. It is meant to work
primarily with version 33.2 of the library and will warn if it
finds another one although that may or may not be fatal. It is
definitely a hack, so it could really scramble the code, but
since it creates the new library in ram: you'll have a chance to
try it before replacing your working library. Programs that use
the Translate() function such as SpeechToy and the SPEAK: device
will receive a harmless error message from Translate() if the
library is corrupt.
USAGE:
assign LIBS: RAM:
TLPatch [-x]
Where "-x" is used to extract the existing exception table
from LIBS:translator.library and write it to ram:except.tbl
adding linefeeds between the exceptions. This can be used if you
wish to start with the original exception table or if you lose
your working copy of except.tbl. Without the "-x" option, TLPatch
will read except.tbl, which it expects to find in ram:, and create
a new translator.library in ram:.
The assignment command is necessary when you first begin the
session if you wish to try the new library, otherwise speech
programs will continue to use the library they found in LIBS: on
disk. Subsequent invocations of TLPatch will free the library
pointer so that your latest edited library is the valid one.
EXCEPTIONS
An exception consists of text on the left of an '=' sign and
the resultant phonemes on the right. The text contains the actual
characters to be converted which are surrounded by square
brackets [], plus any leading or trailing characters whose context
might effect the sound, and possibly substitution symbols which
can match a variety of input strings. Upper-case is used for the
input text and phonemes.
For example:
[AGO] =AHGOW2\
Here all the input characters are translated and because the
input text has a space before and after it, only the word "ago"
will match this exception. If there were no spaces, "lagoon" would
also match, since it contains "ago", which is not desired. This
might be termed a "specific exception" as it fits a specific word.
Numbers such as the '2' in the example above can be added in the
phonetic string to add emphasis to vowels and dipthongs (not
consonants). Values from 1 to 5 are common.
[A]SION=EY3\
There is no leading space so this is a general exception which
will match all words ending "asion". Note that only the "a" part
is translated, the remainder being handled elsewhere.
#:[EA] =IYAH\
These are the tricky ones, being sort of "rule-based" they
match in cases where ANY vowel, consonant, etc. can occupy the
position of its substitution symbol in the input text. They are
tricky because if you create one, there may be words which you
hadn't considered which match. You might also defeat words which
had been correctly translated by creating an unforseen match.
In the example above "idea", "area" or "panacea" all fit.
The advantage of rule-based exceptions is that they keep
the size of the library down by handling so many words.
There are eight substitution symbols that I am aware of,
they are ? ^ % + # : & and @. After studying the library and
experimenting I believe they represent the following values:
? = numbers (0,1,2,3,4,5,6,7,8,9)
e.g. ?[X]?= BAY` - as in "2x4" (the 'X' becomes "by")
^ = consonants (c,f,h,k,p,q,s,x)
e.g. ^[AS]#=EYS\ - as in "CASE"
% = vowels (e)
e.g. [NGL]%=NXGUL\ - as in "ANGLE"
+ = vowels (e,i,y)
e.g. [G]+=J\ - as in "AGE"
# = vowels (a,e,i,o,u,y)
e.g. [BUS]#=BIH3Z\ - as in "BUSY"
: = anything except a,e,i,o,u,y including NULL or SPACE
& = consonants (c,g,j,s,x,z)
e.g. &[T]EN=\ - as in "LISTEN" (the 'T' is silent)
@ = consonants (d,j,l,n,r,s,t,z)
e.g. @[EW]=UW\ - as in "CREW","DEW" or "NEW"
Exceptions always end with either a "\" or "`", I have not
determined what difference it makes although the "\" is by far
the most common.
I've included a simple program called "Translate" to access
the Translate() function. Usage from the CLI is:
Translate <text>
<text> is your input text and can be one word as in:
Translate hello
or a sentence in quotes as in:
Translate "hello world"
Also included is my latest translator.library and exception
table, containing nearly 800 exceptions. Most notably corrections
to numbers as well as month abbreviations (Jan. Feb. etc.). A date
such as "1984" will read "nineteen eighty four" not "one nine eight
four". A number like "27,543" is pronounced "twenty seven thousand
five hundred and forty three" (the comma is important).
TABLE OF PHONEMES
(from RKM)
Vowels -
IY beet | IH bit | EH bet
AE bat | AA hot | AH under
AO talk | UH look | ER bird
OH border | AX about | IX solid
Dipthongs -
EY made | AY hide | OY boil
AW power | OW low | UW crew
Consonants -
R red | L yellow | W away
Y yellow | M men | N men
NX sing | SH rush | S sail
TH thin | F fed | ZH pleasure
Z has | DH then | V very
J judge | CH check | /C loch
/H hole | P put | B but
T toy | D dog | G guest
K could
Special symbols -
DX pity | Q kitten | QX pause
RX car | LX call
Contractions -
UL = AXL | IL = IXL | UM = AXM
IM = IXM | UN = AXN | IN = IXN