Speech::Recognizer::ViaVoice 0.02

Speech::Recognizer::ViaVoice contains Perl wrappers for IBMs ViaVoice speech Recognizer library.

SYNOPSIS

use Speech::Recognizer::ViaVoice;
$| = 1;
if (0 == connectEngine) {
if ( 0 == defineVocab(myTestVocab, [hello,world,quit]) ) {
print please say "hello", "world", or "quit" at each prompt..., "n";

while (0 == startListening) {
print "speak> ";
if (0 == recognize) {
my ($s, $score) = (getWord, getScore);
if (defined($s)) {
printf "%s, score=%dn", $s, $score;
if ($s eq quit) {
exit 0;
}
} else {
printf "not recognized!n";
}
}
stopListening;
}
}

disconnectEngine;
}

This module provides very basic use of IBMs ViaVoice library (http://www-4.ibm.com/software/speech/enterprise/te_3.html) in perl. It allows its user to pass a list of speech commands for recognition in a perl list, connect to the engine, and request recognized spoken words. The most notable feature missing from the current version is the ability to select different ViaVoice users. I plan to add this functionality shortly, but for now you must use the default user.
Consult the ViaVoice documentation for ViaVoice setup. Most will probably just need to run vvstartuserguru, which requires a java runtime environment. Results will be better if you take the time to do a little training to your own voice in this utility.

You can use this perl module for reconition of words from IBMs vocab dictionary for your chosen locale without further preparation. To recognize words not available in the dictionary or to recognize phrases, you need to create a file with phonetic spellings for use by the provided script pronunciations.pl (installed in /usr/local/bin by default). This utility reads a text file whose name is supplied as its sole argument. The file should contain lines with two fields separated by a tab character:

Deborah --- TAB -->D EH B OW R AX
Deborah --- TAB -->D EH B AXR R AX
Axl Rose -- TAB -->AE K S AX L R OW Z

The text on the left is the word or phrase to be recognized, and the text on the right is a phonetic spelling of the format specified in the ViaVoice docs (section 3 of /usr/doc/ViaVoice/bpreadme.txt in my installation). As in the example, a word or phrase might have more than one common pronunciation. You can provide all of them, one on each line.

pronunciations.pl will most likely require super user privileges to write its output to files in the ViaVoice lib directory:

/usr/lib/ViaVoice/vocabs/langs/En_US/pools/