Examples of RecoService entries

The examples below show how to configure RecoService entries for use with:

WebSphere Voice Server Version 5.1. Note that there are differences when using WebSphere Voice Server Version 5.1.1 or 5.1.2 compared to WebSphere Voice Server Version 5.1.3. This is because WebSphere Voice Server Version 5.1.3 supports the use of multiple recognition languages on a single machine. The examples also show that the InitTechnologyString keyword is not required when using any of these versions of WebSphere Voice Server speech technology.
Nuance Recognizer V9.0.3

Configuring for use with WebSphere Voice Server Version 5.1.1 or 5.1.2

In this example, two WebSphere Voice Server languages are needed: US English and Latin American Spanish. US English is installed on the WebSphere Voice Server machine with host name wvsenglish.demo.ibm.com. Latin American Spanish is installed on the WebSphere Voice Server machine with host name wvslaspanish.demo.ibm.com. The RecoService entry in the default.cff file would include the following values:

RecoService=WVS_Recoen_US
 PluginClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvsenglish.demo.ibm.com/media/recognizer
 RecoType=Recoen_US
;
RecoService=WVS_Recoes_MX
 PluginClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvslaspanish.demo.ibm.com/media/recognizer
 RecoType=Recoes_MX
;

On a Blueworx Voice Response machine with more than one ethernet adapter you also need to define the local IP address to be used for the media streaming by specifying a MediaIPAddress value for the InitSessionString parameter:

RecoService=WVS_Recoen_US
 PluginClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvsenglish.demo.ibm.com/media/recognizer, MediaIPAddress=9.20.123.456
 RecoType=Recoen_US
;
RecoService=WVS_Recoes_MX
 PluginClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvslaspanish.demo.ibm.com/media/recognizer, MediaIPAddress=9.20.123.456
 RecoType=Recoes_MX
;

In each case, this speech recognition configuration would be reflected in the NodeName configuration entry as follows:

NodeName=VRNode1
 Enabled=yes
 NodeDefLocale=en_US
 VRNode=yes
 RecoService=WVS_Recoen_US
 RecoDefinition=en_US,Recoen_US
 RecoService=WVS_Recoes_MX
 RecoDefinition=es_MX,Recoes_MX
;

Configuring for use with WebSphere Voice Server Version 5.1.3

WebSphere Voice Server Version 5.1.3 allows a single machine to support multiple speech recognition languages. This means that default.cff needs to contain only a single reference to a WebSphere Voice Server machine for each speech recognition resource type.

In this example, a machine with host name wvs.demo.ibm.com has installed US English and Latin American Spanish. The RecoService entry in the default.cff file would include the following values:

RecoService=RecoAll
 PlugInClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvs.demo.ibm.com/media/recognizer
 RecoType=Recolan_All
;

RecoService=RecoAll
 PlugInClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
 InitSessionString=URI=rtsp://wvs.demo.ibm.com/media/recognizer, MediaIPAddress=9.20.123.456
 RecoType=Recolan_All
;

In each case, this speech recognition configuration would be reflected in the NodeName configuration entry as follows:

NodeName=VRNode1
 Enabled=yes
 NodeDefLocale=en_US
 VRNode=yes
 RecoService=RecoAll
 RecoDefinition=*,Recolan_All
;

Using load-balancing with WebSphere Voice Server

For larger deployments handling more than one trunk of calls, it may be necessary to use more than one speech server to process text-to-speech and speech recognition requests. To do this a load-balancing application such as WebSphere Edge Server is used to distribute text-to-speech and speech recognition requests to two or more speech servers. All text-to-speech and speech recognition initialization requests are sent to a load-balancing address. The load balancer receives these requests and forwards them to one of a cluster of speech servers. For more information, refer to the WebSphere Voice Server information center topic “Multiple machine topology”.

Note: WebSphere Edge Server V5.1 is bundled with WebSphere Voice Server V5.1.3.

For instructions on installing WebSphere Edge Server, refer to the WebSphere Voice Server information center topic “Installing the WebSphere Edge Component: Load Balancer”.

Configuring for use with Nuance Recognizer V9.0.3

Blueworx Voice Response uses MRCP V1.0 to connect to voice servers. Nuance Speech Server V5 provides an MRCP V1.0 interface to text-to-speech and speech recognition components. The default RTSP port number to use to connect to the Nuance Speech Server must match the Nuance Server configuration. By default this value is 4900.

An example of the VRBE RecoService entry configuration settings required for speech recognition in configuration file /var/dirTalk/DTBE/native/aix/default.cff are shown below:

RecoService=Reco_GB
  PlugInClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
  InitSessionString=URI=rtsp://1.23.45.678:4900/media/speechrecognizer
  RecoType=Recoen_GB
;

Using load-balancing with Nuance Recognizer V9.0.3

For larger deployments handling more than one trunk of calls, it may be necessary to use more than one speech server to process text-to-speech and speech recognition requests. To do this a load-balancing application such as WebSphere Edge Server is used to distribute text-to-speech and speech recognition requests to two or more speech servers. All text-to-speech and speech recognition initialization requests are sent to a load-balancing address. The load balancer receives these requests and forwards them to one of a cluster of speech servers. To use Nuance speech servers in this way, a number of changes to Blueworx Voice Response and Nuance configuration need to be made.

An example of the Java and VoiceXML environment RecoService entry configuration settings required for speech recognition in configuration file default.cff when using a load-balanced systems are shown below:

RecoService=Reco_GB
  PlugInClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
  InitSessionString=URI=rtsp://1.23.45.678:554/media/speechrecognizer
  RecoType=Recoen_GB
;

With load-balanced systems, port 554 is used instead of 4900. The changes required to the Nuance configuration file NSSserver.cfg are:

Variable	Value
`server.transport.bindrtptoip`	The IP address of the speech server
`server.rtp.strictSdpMediaPortUse`	0
`server.mrcp1.transport.port`	554

Speech recognition requests are forwarded to one of a number of Nuance voice servers using the same port number (554). When a speech recognition session has been set up on a Nuance voice server and Blueworx Voice Response has received a response, speech recognition traffic passes to and fro directly (not through the load balancer machine).

Configuring for use with Loquendo ASR V7.9

Blueworx Voice Response uses MRCP V1.0 to connect to voice servers. Loquendo 7 provides an MRCP V1.0 interface to text-to-speech and speech recognition components. The default RTSP port number to use to connect to the Loquendo Speech Server must match the Loquendo Server configuration. By default this value is 554.

An example of the VRBE RecoService entry configuration settings required for speech recognition in configuration file /var/dirTalk/DTBE/native/aix/default.cff are shown below:

RecoService=Reco_GB
  PlugInClass=com.ibm.telephony.directtalk.mrcp.MRCPReco
  InitSessionString=URI=rtsp://1.23.45.678:554/recognizer
  RecoType=Recoen_GB
;

In the Loquendo Management Console, the following two configuration settings need to be set:

Configuration > Advanced > MRCPv1Server > nlsmlResult > enableWordInputElements:

disabled

Configuration > Advanced > MRCPv1Server > speechRecognition > lasrDefaultTagFormat:

SISR-semantics (2)

To use remote DTMF grammars, the following two configuration settings also need to be set:

Configuration > Basic > MRCPv1Server > dtmfCodec:

Configuration > Basic > MRCPv1Server > dtmfCodecAlwaysOffered:

enabled (1)

After changing configuration settings, it may be necessary to restart Loquendo.