NAME
speak - word to voice translator
SYNOPSIS
speak [ -efpsv ] [ vocabulary [ output ] ]
DESCRIPTION
Speak turns a stream of words into utterances and outputs
them to a voice synthesizer, or to the specified output. It
has facilities for maintaining a vocabulary. It receives,
from the standard input
- working lines: text of words separated by blanks
- phonetic lines: strings of phonemes for one word pre-
ceded and separated by commas. The phonemes may be
followed by comma-percent then a `replacement part' -
an ASCII string with no spaces. The phonetic code is
given in vs (V).
- empty lines
- command lines: beginning with !. The following command
lines are recognized:
!r file replace coded vocabulary from file
!w file write coded vocabulary on file
!p print phonetics for working word
!l list vocabulary on standard output with pho-
netics
!c word copy phonetics from working word to specified
word
!d print decomposition of working word into sub-
strings
!f n turn off (or on) English preprocessing rule
number n (see listing for meaning of n)
Each working line replaces its predecessor. Its first word
is the `working word'. Each phonetic line replaces the pho-
netics stored for the working word. In particular, a pho-
netic line of comma only deletes the entry for the working
word. Each working line, phonetic line or empty line causes
the working line to be uttered. The process terminates at
the end of input.
Unknown words are pronounced by rules, and failing that, are
spelled. For the builtin part of the rules, see the refer-
ence. Spelling is done by taking each character of the
word, prefixing it with `*', and looking it up. Unspellable
words burp.
Words not found verbatim in the vocabulary are pronounced
piecewise. First the word is bracketed by sharps: `#...#'.
The vocabulary is then searched for the longest fragment
that matches the beginning of the word. The phonetic part
of the phonetic string is uttered, and the matched fragment
1
is replaced by the replacement part of the phonetic string,
if any. The process is repeated until the word is
exhausted. A fragment is entered into the vocabulary as a
working word prefixed by `%'.
Speak is initialized with a coded vocabulary stored in file
/usr/lib/speak.m. The vocabulary option substitutes a dif-
ferent file for /usr/lib/speak.m. Other vocabularies, to be
used with option -e, exist in /usr/vs/latin.m and
/usr/vs/polish.m.
A set of single letter options may appear in any order pre-
ceded by -. Their meanings are:
e suppress English preprocessing
f equivalent to `f1, f2,...'
p suppress pronunciation by rule
s suppress spelling
v suppress voice output
The following input will reconstitute a coded vocabulary,
`speak.m', from an ascii listing, `speak.v', that was cre-
ated using !l.
(cat speak.v; echo !w speak.m) | speak -v /dev/null
FILES
/usr/lib/speak.m
SEE ALSO
M. D. McIlroy, ``Synthetic English Speech by Rule,'' Comput-
ing Science Technical Report #14, Bell Laboratories, 1973
vs (V), vs (IV)
BUGS
Excessively long words cause dumps.
Space is not reclaimed from changed entries; use !w and !r
to effect reclamation.
!p doesn't always work as advertised.
2