Re: Asterisk+UniMRCP+Nuance: Valid NLSML but SPEECH_TEXT is null

[Hernan P.]

Note - I am using:

* Asterisk v.1.8.15

* Nuance Recognizer v.10.2, NSS v.6.2

* UniMRCP v.1.0.0

Thanks,

HP

On Friday, August 10, 2012 5:18:55 PM UTC-5, Hernan P. wrote:

Hi,

We are trying to integrate Asterisk and Nuance with UniMRCP. After a successful installation, we are using the sample dialplan extension that comes with UniMRCP to test speech recognition in Asterisk:

exten => ts,1,Answer()

exten => ts,2,SpeechCreate()

exten => ts,3,SpeechLoadGrammar(digit,/usr/local/unimrcp/data/grammar.xml)

exten => ts,4,SpeechActivateGrammar(digit)

exten => ts,5,SpeechBackground(hello-world,20)

exten => ts,6,GotoIf($["${SPEECH(results)}" = "0"]?7:9)

exten => ts,7,Playback(vm-nonumber)

exten => ts,8,Goto(5)

exten => ts,9,Verbose(1,The recognized input is ${SPEECH_TEXT(0)})

exten => ts,n,Verbose(1,The score is ${SPEECH_SCORE(0)})

exten => ts,n,Verbose(1,The matched grammar is ${SPEECH_GRAMMAR(0)})

exten => ts,n,SpeechDeactivateGrammar(digit)

exten => ts,n,SpeechUnloadGrammar(digit)

exten => ts,n,SpeechDestroy()

exten => ts,n,Hangup()

The problem: The variable SPEECH_TEXT is always empty. The SPEECH_SCORE and SPEECH_GRAMMAR get valid values though. No errors show up in the logs.

I looked at the logs and Asterisk is getting the recognition NLSML from Nuance server:

<?xml version='1.0'?><result><interpretation grammar="session:digit" confidence="0.78"><input mode="speech">two</input><instance><SWI_literal>two</SWI_literal><SWI_grammarName>session:digit</SWI_grammarName><SWI_meaning>{SWI_literal:two}</SWI_meaning></instance></interpretation></result>

However, the log then shows:

NOTICE[20673] res_speech_unimrcp.c: Interpreted instance:none score:78 grammar:digit

I compared to the result.xml that comes with UniMRCP:

<?xml version="1.0"?>

<result>

  <interpretation grammar="session:request1@form-level.store" confidence="97">

    <instance>one</instance>

    <input mode="speech">one</input>

  </interpretation>

</result>

...and I noticed that Nuance's response has more elements under <instance>.

So I thought that UniMRCP may not be parsing the NLSML response correctly. This is what I found in res-speech-unimrcp.c source file:

/** \brief Try to get result */

struct ast_speech_result* uni_recog_get(struct ast_speech *speech)

{

[...]

 interpret = nlsml_first_interpret_get(doc);

        if(interpret) {

                apr_xml_elem *instance;

                apr_xml_elem *input;

                /* Get instance and input */

                nlsml_interpret_results_get(interpret,&instance,&input);

                if(instance && input) {

                        const char *confidence;

                        const char *grammar;

                        speech->results = ast_calloc(sizeof(struct ast_speech_result), 1);

                        speech->results->text = NULL;

                        speech->results->score = 0;

                        if(instance->first_cdata.first) {

                                speech->results->text = strdup(instance->first_cdata.first->text);

                        }

                        confidence = nlsml_input_attrib_get(instance,"confidence",TRUE);

                        if(confidence) {

                                if(uni_speech->mrcp_event->start_line.version == MRCP_VERSION_2) {

                                        speech->results->score = (int)(atof(confidence) * 100);

                                }

                                else {

                                        speech->results->score = atoi(confidence);

                                }

                        }

                        grammar = nlsml_input_attrib_get(interpret,"grammar",TRUE);

                        if(grammar) {

                                grammar = strchr(grammar,':');

                                if(grammar && *grammar != '\0') {

                                        grammar++;

                                        speech->results->grammar = strdup(grammar);

                                }

                        }

                        ast_log(LOG_NOTICE, "Interpreted instance:%s score:%d grammar:%s\n",

                                speech->results->text ? speech->results->text : "none",

                                speech->results->score,

                                speech->results->grammar ? speech->results->grammar : "none");

                        ast_set_flag(speech,AST_SPEECH_HAVE_RESULTS);

                }

        }

        return speech->results;

}

...From what I understand, it looks like it is trying to find the text as the first element in <instance>, and maybe that is why the string is always empty?

The question is: What is the correct way of fixing this problem? Changing res-speech-unimrcp.c code? or trying to make nuance send an xml that UniMRCP is expecting? Or is the root of this problem somewhere else and I am missing something? Anyone with previous experience integrating with Nuance that could point me in the right direction?

Thanks a lot in advance,

Hernan Palombo

[Hernan P.]

UPDATE - I solved this problem by installing UniMRCP and Asterisk Connector from svn source.

Thanks,

Hernan

[Arsen Chaloyan]

Hi Hernan,

Thanks for making the final resolution clear to us.

All: please use the latest trunk, it contains many fixes which are not
included in the last released packages.

>>> <interpretation grammar="session:requ...@form-level.store"

> --
> You received this message because you are subscribed to the Google Groups
> "UniMRCP" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/unimrcp/-/WW4r2xLlbpYJ.
>
> To post to this group, send email to uni...@googlegroups.com.
> To unsubscribe from this group, send email to
> unimrcp+u...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/unimrcp?hl=en.



--
Arsen Chaloyan
Author of UniMRCP
http://www.unimrcp.org