Posted: Wed May 16, 2007 10:51 pm Post subject: [asterisk-speech-rec] How to enable N-Best decoding?
Scott (and anyone else interested in n-best),
I apologize for the delay in this response, but we've been working with
Digium on troubleshooting some issues related to n-best using the
LumenVox Speech Engine on Asterisk. It looks like we have them all
straightened out, and the latest version of the Asterisk-LumenVox
connector bridge is available on our FTP site (as always drop me a line
off-list if you need the link).
To activate N-best, use Set(SPEECH_RESULTS_TYPE()=nbest) after
SpeechCreate.
You can then check results like normal, except you specify different
results using ${SPEECH_TEXT(X/Y)}, where X is the number of the result
to check (starting at zero) and Y is the interpretation.
So the first interpretation for the first n-best result is 0/0; e.g.
${SPEECH_TEXT(0/0)}. If there is more than one semantic interpretation,
you would check 0/1. The second n-best result would be 1/0, etc.
Likewise you would use ${SPEECH_SCORE(X/Y)} in the same fashion to check
confidence scores.
Here's a very basic dial plan and grammar that should illustrate the
concepts. If you use the grammar and say a valid choice, you will almost
certainly get back multiple n-best results, since all the words are very
similar.
exten => 1,1,Answer
exten => 1,n,Wait(1)
exten => 1,n,SpeechCreate
exten => 1,n,Set(SPEECH_RESULTS_TYPE()=nbest)
exten => 1,n,SpeechActivateGrammar(test)
exten => 1,n,SpeechBackground(beep)
exten => 1,n,Verbose(1,Number of results: ${SPEECH(results)})
exten => 1,n,Verbose(1,First result is ${SPEECH_TEXT(0/0)} with a
confidence score of ${SPEECH_SCORE(0/0)})
exten => 1,n,Verbose(1,Second result is ${SPEECH_TEXT(1/0)} with a
confidence score of ${SPEECH_SCORE(1/0)})
Test grammar:
#ABNF 1.0 UTF-8;
language en-US;
mode voice;
tag-format <semantics/1.0>;
I'm running the Lumenvox 7.5 engine, the Lumenvox 1.4 connector, and
Asterisk 1.4.3. All is well, except that the N-Best feature
does not
seem to work. As I understand it, this feature allows the
recogniser to return multiple possibilities for the decoded
recognition.
What do I have to do to enable this? An excerpt from the
DecodeServerLog.txt file shows these interesting lines before
each decode. Perhaps the "NO_OOV" is important?
04/30/2007
09:28:12,860,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0] Begin
Decode Using Context-Free Grammar
04/30/2007 09:28:12,861,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0]
Loading Context Free Gramamr
04/30/2007 09:28:12,863,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0]
Retrieving words
04/30/2007
09:28:12,864,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0] Word
retrieval time: 0
04/30/2007
09:28:12,877,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0] nbest
will not be used on this decode run
04/30/2007
09:28:12,877,0,[AmericanEnglish][MODEL_LOW:ThreadNdx:0] Begin
Core Decode Port
Posted: Thu May 17, 2007 4:00 pm Post subject: [asterisk-speech-rec] How to enable N-Best decoding?
Hi Stephen:
Quote:
connector bridge is available on our FTP site (as
always drop me a line
off-list if you need the link).
Yes, if could send me the link of the new connector
bridge (for Fedora 6) that would be great.
Also, I've been finding that a lot of times the Speech
engine takes a long time (on the order of 1 or a
several seconds) to recognize speech. This happens on
my box with simple grammars (such as the Pizza demo)
or even more complex ones (the liveWeather demo I
submitted for the Lumenvox competition). Is this
normal? And if not, do you have any tips on tracking
this down?
Thanks in advance,
Adrian
--- Stephen Keller <StephenKeller@LumenVox.com> wrote:
Quote:
Scott (and anyone else interested in n-best),
I apologize for the delay in this response, but
we've been working with
Digium on troubleshooting some issues related to
n-best using the
LumenVox Speech Engine on Asterisk. It looks like we
have them all
straightened out, and the latest version of the
Asterisk-LumenVox
connector bridge is available on our FTP site (as
always drop me a line
off-list if you need the link).
To activate N-best, use
Set(SPEECH_RESULTS_TYPE()=nbest) after
SpeechCreate.
You can then check results like normal, except you
specify different
results using ${SPEECH_TEXT(X/Y)}, where X is the
number of the result
to check (starting at zero) and Y is the
interpretation.
So the first interpretation for the first n-best
result is 0/0; e.g.
${SPEECH_TEXT(0/0)}. If there is more than one
semantic interpretation,
you would check 0/1. The second n-best result would
be 1/0, etc.
Likewise you would use ${SPEECH_SCORE(X/Y)} in the
same fashion to check
confidence scores.
Here's a very basic dial plan and grammar that
should illustrate the
concepts. If you use the grammar and say a valid
choice, you will almost
certainly get back multiple n-best results, since
all the words are very
similar.
exten => 1,1,Answer
exten => 1,n,Wait(1)
exten => 1,n,SpeechCreate
exten => 1,n,Set(SPEECH_RESULTS_TYPE()=nbest)
exten => 1,n,SpeechActivateGrammar(test)
exten => 1,n,SpeechBackground(beep)
exten => 1,n,Verbose(1,Number of results:
${SPEECH(results)})
exten => 1,n,Verbose(1,First result is
${SPEECH_TEXT(0/0)} with a
confidence score of ${SPEECH_SCORE(0/0)})
exten => 1,n,Verbose(1,Second result is
${SPEECH_TEXT(1/0)} with a
confidence score of ${SPEECH_SCORE(1/0)})
Test grammar:
#ABNF 1.0 UTF-8;
language en-US;
mode voice;
tag-format <semantics/1.0>;
On
> Behalf Of Scott Stingel
> Sent: Monday, April 30, 2007 9:38 AM
> To: asterisk-speech-rec@lists.digium.com
> Subject: [asterisk-speech-rec] How to enable
N-Best decoding?
>
> Hello-
>
> I'm running the Lumenvox 7.5 engine, the Lumenvox
1.4 connector, and
> Asterisk 1.4.3. All is well, except that the
N-Best feature
> does not
> seem to work. As I understand it, this feature
allows the
> recogniser to return multiple possibilities for
the decoded
> recognition.
>
> What do I have to do to enable this? An excerpt
from the
> DecodeServerLog.txt file shows these interesting
lines before
> each decode. Perhaps the "NO_OOV" is important?
>
> 04/30/2007
>
Begin
> Core Decode Port
>
> Thanks for any help,
> Scott Stingel
>
> www.evtmedia.com
>
> _______________________________________________
> --Bandwidth and Colocation provided by
Easynews.com --
>
> asterisk-speech-rec mailing list
> To UNSUBSCRIBE or update options visit:
>
You can post new topics in this forum You can reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum