RC8650 datasheet(39/48 Pages) ETC

39 / 48 page

RC8650 VOICE SYNTHESIZER

RC SYSTEMS

$R(H)=

which states that h after initial r is silent, as in the word rhyme (the

context token represents any non-alphabetic character, such as a

space between words; see Table 4.1).

Punctuation, numbers, and most other characters can be redeﬁned

with exceptions as well:

(5)=S I NG K O

(Spanish ﬁve)

(CHR$)=K EH R IX K T ER

(Basic function)

THE TRANSLATION ALGORITHM

In order to better understand how an exception dictionary works, it

is helpful to understand how the DoubleTalk text-to-speech engine

processes text.

Algorithms within the DoubleTalk engine analyze input text a charac-

ter at a time, from left to right. A list of pronunciation rules is searched

sequentially for each character until a rule is found that matches the

character in the correct position and context. The algorithm then

passes over the input character(s) bracketed in the rule (the text frag-

ment), and assigns the pronunciation given by the right side of the

rule to them. This process continues until all of the input text has been

converted to phonetic sounds.

The following example illustrates how the algorithm works by translat-

ing the word receive.

The algorithm begins with the letter r and searches the R pronunciation

rules for a match. The ﬁrst rule that matches is

$(RE)^#=R IX, be-

cause the r in receive is an initial r and is followed by an e, a consonant

(c), and a vowel (e). Consequently, the text fragment re receives the

pronunciation

R IH, and the scan moves past re to the next character:

receive. (E is not the next scan character because it occurred inside

the parentheses with the r; the text fragment re as a whole receives

the pronunciation

R IX)

The ﬁrst match among the C rules is

(C)+=S, because c is followed

by an e, i, or y. C thus receives the pronunciation

S, and processing

continues with the second e: receive.

(EI)=IY is the ﬁrst rule to match the second e, so ei receives the

sound

IY.Processingresumesatthecharacterreceive,whichmatch-

es the default V rule,

(V)=V.

The ﬁnal e matches the rule

#:(E)$=, which applies when e is ﬁnal

and follows zero or more consonants and a vowel. Consequently, e

receives no sound and processing continues with the following word

or punctuation, if any. Thus, the entire phoneme string for the word

receive is

R IX S IY V.

RULE PRECEDENCE

Since DoubleTalk uses its translation rules in a sequential manner, the

position of each exception relative to the others must be carefully con-

sidered. For example, consider the following pair of exceptions:

(O)+=OW

(O)=UW

The ﬁrst exception states that o followed by e, i, or y is to be pro-

nounced

OW, the o-sound in boat. The second exception does not

place any restriction on what must come before or after o, so o in

any context will receive the

UW pronunciation. If the exceptions were

reversed, the

(O)+ exception would never be reached because the

(O) exception will always match o in any context. In general, tightly-

defined exceptions (those containing many context restrictions)

should precede loosely-deﬁned exceptions (those with little or no

context deﬁnitions).

(RAT)=R AE T

(RATING)=R EY T IH NG

(R)=R

This is an example of how not to organize exceptions. The exception

(RATING) will never be used because (RAT) will always match ﬁrst.

According to these exceptions, the word rating would be pronounced

“rat-ing.”

It can be beneﬁcial to group exceptions by the ﬁrst character of the

text fragments, that is, all of the A exceptions in one group, all the B

exceptions in a second group, and so on. This gives an overall cleaner

appearance, and can prove to be helpful if the need arises to trouble-

shoot any problems in your dictionary.

TEXT NOT MATCHED BY THE DICTIONARY

It is possible that some input text may not match anything in a diction-

ary, depending on the nature of the dictionary. For example, if a diction-

ary was written to handle unusual words, only those words would be

included in the dictionary. On the other hand, if a dictionary deﬁned

the pronunciation for another language, it would be comprehensive

enough to handle all types of input. In any case, if an exception is not

found for a particular character, the English pronunciation will be given

to that character according to the built in pronunciation rules.

Generally, the automatic switchover to the built in rules is desirable

if the dictionary is used to correct mispronounced words, since by

deﬁnition the dictionary is deﬁning exceptions to the built in rules. If

the automatic switchover is not desired, however, there are two ways

to prevent it from occurring. One way is to end each group of excep-

tions with an unconditional exception that matches any context. For

example, to ensure that the letter “a” will always be matched, end the

A exception group with the exception

(A)=pronunciation. This tech-

nique works well to ensure matches for speciﬁc characters, such as

certain letters or numbers.

If the exception dictionary is to replace the built in rules entirely, end

the dictionary with the following exception:

()=

This special exception causes unmatched characters to be ignored

(receive no sound), rather than receive the pronunciation deﬁned by

the built in rules.

EFFECT ON PUNCTUATION

RC8650 датащи(PDF) 39 Page - List of Unclassifed Manufacturers

RC8650 датащи(HTML) 39 Page - List of Unclassifed Manufacturers

Html Pages

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

датащи скачать

ссылки URL