[asterisk-speech-rec] How to enable N-Best decoding?

[Stephen Keller]

> A few months ago you gave a good description on how to use 
> the N-Best feature in Lumenvox. I think I understand the 
> difference between ${SPEECH_TEXT(0/0)} and 
> ${SPEECH_TEXT(1/0)}. That was made clear from your example 
> with ($she | $sea | $he | $bee). 
> Unfortunately, I'm not sure I understand the difference 
> between ${SPEECH_TEXT(0/0)} and
> ${SPEECH_TEXT(0/1)}.   Do you have an example you
> could send illustrating the difference? Basically, I'm not 
> sure what you meant when you said: 
> 
> > If there is more than one semantic interpretation, you would check 
> > 0/1.

This happens when the same input can match multiple rules, giving the
Engine two or more valid parses of a grammar.

A common example is a call router where you just allow first name
matches, but have two people with the same first name:

	root $name;

	$billjones = bill [jones] {out = "bill jones"};
	$billsmith = bill [smith] {out = "bill smith"};

	$name = ($billjones | $billsmith) {out = rules.latest()};

If a caller just says "Bill" then you have two distinct parses, each
with a different semantic interpretation. In the first parse, the
interpretation is "bill jones" and in the second the interpretation is
"bill smith".

> Also, the ${SPEECH(results)} variable is used to determine 
> the total number of N-Best results. Is there an equivalent 
> variable that can be used to check for the number of 
> "semantic interpretations"?

I'm not sure. If one of the Digium folks could answer this, I would be
grateful.

Thanks,

Stephen Keller
LumenVox Support
P: 877-977-0707, just say "Support"
F: 858-707-7072
Support at LumenVox.com
www.LumenVox.com



Winner "Best Innovation in Speech Recognition"
AVIOS SpeechTEK Award