Configuring MRCP clients

This topic describes the configuration of resources on the Speech Server using Management Station. The purpose of this configuration is to match Nuance Speech Server settings to the expectations of the MRCP client.

Note: Speech Server supports both MRCPv2 and MRCPv1; however, Nuance recommends using MRCPv2. For example, Dragon Voice does not support MRCPv1.

Although Speech Server generally works well with the installation defaults, your speech browser (and its MRCP client) might require non-default values. System administrators must review the defaults to determine appropriate values, and if necessary modify settings.

In general, these parameters are not modified for individual applications. Instead, application developers use a session.xml. For details, see Configuring application sessions.

Configuring network security

Speech Server supports enhanced security for communication with MRCPv2 clients. Use Transport Layer Security (TLS) to encrypt SIP and MRCP requests and responses between the MRCP client and Speech Server so that requests and responses remain invisible to outside observers.

Speech Server supports TLS versions 1.0, 1.1, and 1.2.

Note: You can also use Secure Real-time Transport Protocol (SRTP) to encrypt the audio channel. See Setting audio channel security.

These parameters configure TLS for Speech Server:

Parameter

Description

Value

server.mrcp2.sip.transport.tls.port

Specifies the SIP TLS port to use for the application server.

An available port number.

DEFAULT: 5061

server.mrcp2.transport.tls.port

Specifies the MRCPv2 TLS port to use for the application server.

An available port number.

DEFAULT: 6076
server.callLog.tls.port

Specifies the listening TLS port for clients to communicate with call log server.

Listening TLS port for clients to communicate with call log server.

Integer. An available port number for TLS.

DEFAULT: 10102
server.tls.tlsVersion

Specifies the allowed versions of the TLS protocol.

Integer. One of the following:

  • 0 (TLS v1.0 only)
  • 1 (TLS v1.1 and above)
  • 2 (TLS v1.2 and above)

DEFAULT: 0

You can enable two-way peer authentication with web servers with these Speech Server parameters:

Parameter

Description

Value

server.tls.caCertificatesFile

Specifies the file containing one or more sequential PEM-encoded public CA certificates.

String

DEFAULT: (none)

server.tls.certificateFile

Specifies the PEM-encoded certificate for Speech Server.

String

DEFAULT: (none)

server.tls.privateKeyFile

Specifies the PEM-encoded private key for Speech Server.

String

DEFAULT: (none)

server.inet.tls.verify

Enables peer authentication.

Boolean

DEFAULT: 0

server.tls.verifyDepth

Limits the depth of the certificate chain for validation.

Integer

DEFAULT: 2 (accommodates one intermediate CA)

Set these Speech Server parameters to require the highest level of security:

Parameter

Description

Value

server.mrcp2.sip.transport.tls.useStrongestCipherSuite

Specifies the encryption ciphers on the TLS port. Requires use of only the strongest TLS ciphers.

Integer. Must be 0 (accept weak ciphers), 1 (accept strong ciphers only), or 2 (accept cipher suites based on the DH key exchange method with GCM mode only).

DEFAULT: 0 (accept weak ciphers)

server.mrcp2.transport.tls.useStrongestCipherSuite

Specifies the encryption ciphers on the TLS port. Requires use of only the strongest TLS ciphers.

Integer. Must be 0 (accept weak ciphers), 1 (accept strong ciphers only), or 2 (accept cipher suites based on the DH key exchange method with GCM mode only).

DEFAULT: 0 (accept weak ciphers)
server.callLog.useStrongestCipherSuite

Specifies the encryption ciphers on the TLS port. Requires use of only the strongest TLS ciphers for communication with the call log server.

Integer. Must be 0 (accept weak ciphers), 1 (accept strong ciphers only), or 2 (accept cipher suites based on the DH key exchange method with GCM mode only).

DEFAULT: 0 (accept weak ciphers)

Configuring recognition resources

This section describes the configuration of speech recognition resources on Speech Server in Management Station, including for selective barge-in and magic word.

Note: The Krypton recognition engine does not support hotword mode recognition, including selective barge-in and magic word.

Set these parameters if your voice browser uses MRCPv2. (If your browser uses MRCPv1, it ignores these parameters.) You can control several types of activity:

Parameter Description Value

Audio processing

server.mrcp2.audioengine.audioThreadNumber

Specifies the number of audio threads used to fetch audio and feed it to the recognizer and recorder.

Integer: 1–INT_MAX.

DEFAULT: 20

Endpointer usage

server.mrcp2.osrspeechrecog.endpointer

Controls use of the endpointer.

Boolean

DEFAULT: 1 (external endpointer)

Enable/disable use of cookies with Internet fetches

server.mrcp2.osrspeechrecog.cookie.enable

Enables the use of cookies for retrieving files.

Boolean

DEFAULT: 0

Responses to various recognitions

server.mrcp2.osrspeechrecog.internalBargein

Sends a BARGE-IN-OCCURRED event directly from the recognizer to Vocalizer to quickly stop a prompt.

Boolean

DEFAULT: 0 (disabled)

server.mrcp2.osrspeechrecog.startOfInputOnDTMF

Send a START-OF-INPUT event on DTMF input.

Boolean

DEFAULT: 0 (disabled)

server.mrcp2.osrspeechrecog.startOfInputOnHotword

Send a START-OF-INPUT event each time new candidate speech is detected in a hotword mode recognition.

Boolean

DEFAULT: 0 (disabled)
server.mrcp2.osrspeechrecog.hotwordSuppression

Prevents a hotword from being included in recordings.

Integer. 0 (hotword is not suppressed) or 1 (hotword is suppressed)

DEFAULT: 0 (hotword is not suppressed)

Result formats

server.mrcp2.osrspeechrecog.mrcpdefaults.VSP.server.
osrspeechrecog.result.mediatype

Specifies the media type of the recognition result returned to the application.

MIME media type supported.

DEFAULT: application/x-vnd.speechworks.emma+xml;strictconfidencelevel=1;mrcpv=2.06

server.mrcp2.osrspeechrecog.mrcpdefaults.VSP.server.
osrspeechrecog.result.sendnomatch

Return the recognition result to the MRCP client even when the confidence is low.

Boolean

DEFAULT: false (does not return the result of low confidence recognitions)

Set these parameters if your voice browser uses MRCPv1. (If your browser uses MRCPv2, it ignores these parameters.) You can control several types of activity:

Parameter Description Value

Audio processing

server.mrcp1.osrspeechrecog.audioBufferSize

Specifies the size of the audio buffer.

The range of values depends on the audio type and the sampling rate. Assuming a sampling rate of 8kHz:

ulaw, alaw: 0–500 milliseconds.

L16: 0–250 milliseconds.

The maximum buffer size is equivalent to 4000 bytes.

DEFAULT: 100 (milliseconds)

server.mrcp1.osrspeechrecog.audioThreadNumber

Specifies the number of threads used to fetch audio and feed it to the recognizer.

Integer: 1–INT_MAX.

DEFAULT: 10

Endpointer usage

server.mrcp1.osrspeechrecog.endpointer

Controls use of the endpointer.

Boolean

DEFAULT: 1 (external endpointer)

Enable/disable use of cookies with Internet fetches

server.mrcp1.osrspeechrecog.cookie.enable

Enables the use of cookies for retrieving files.

Boolean

DEFAULT: 0 (no cookies)

Responses to various recognitions

server.mrcp1.osrspeechrecog.internalBargein

Sends a BARGE-IN-OCCURRED event directly from the Recognizer to Vocalizer to quickly stop a prompt.

Boolean

DEFAULT: 0 (disabled)

server.mrcp1.osrspeechrecog.startOfSpeechOnDTMF

Send a START-OF-SPEECH event on DTMF input.

Boolean

DEFAULT: 1 (enabled)

server.mrcp1.osrspeechrecog.startOfSpeechOnHotword

Send a START-OF-SPEECH event each time new candidate speech is detected in a hotword mode recognition.

Boolean

DEFAULT: 0 (disabled)

Result formats

server.mrcp1.osrspeechrecog.nlsml.encoding

Inserts an XML header before the NLSML results.

Character encoding type

DEFAULT: ISO-8859-1

server.mrcp1.osrspeechrecog.result.mediatype

Specifies the media type of the recognition result returned to the application.

Media type

DEFAULT: application/x-vnd.speechworks.emma+xml;strictconfidencelevel=1

server.mrcp1.osrspeechrecog.startOfSpeechOnDTMF

Send a START-OF-SPEECH event on DTMF input.

Boolean

DEFAULT: 1 (enabled)

An application can use the hotword mode to support two Nuance-specific barge-in modes: selective barge-in and magic word. These modes enable the application to recognize a specific speech or DTMF sequence and ignore anything else.

  • Selective barge-in prevents accidental interruption by allowing applications to define a small set of key words (to be spoken by callers) that trigger barge-in. An application that supports selective barge-in always listens for commands, whether the caller is speaking or listening to prompts.
  • Magic word is identical to selective barge-in except that it also rejects candidates that are too short or long.

The application and must co-ordinate the endpointer and the recognizer, using the parameters swiep_mode and swirec_barge_in_mode. Both resources must be set to compatible modes for any given recognition:

swiep_mode

swirec_barge_in_mode

begin_only (default)

normal

magic_word

magic_word

selective_barge_in

selective_barge_in

You can also use the following parameters to control barge-in modes:

Parameter

Description

Value

swiep_magic_word_max_msec

Specifies the maximum duration of a magic word candidate for recognition.

Integer: milliseconds. Minimum is 0; there is no maximum.

DEFAULT: 800 (milliseconds)

swiep_magic_word_min_msec

Specifies the minimum duration of a magic word candidate for recognition.

Integer: 0– swiep_magic_word_max_msec milliseconds.

DEFAULT: 200 (milliseconds)

swirec_magic_word_conf_thresh

Specifies the confidence threshold for recognition results computed while the magic_word mode is active.

Integer: 0–999.

DEFAULT: 500

swirec_selective_barge_in_conf_thresh

Specifies the confidence threshold for recognition results computed while the selective_barge_in mode is active.

Integer: 0–999.

DEFAULT: 500

incompletetimeout

Specifies the duration of silence to determine that callers have finished speaking.

Integer: 0–INT_MAX (milliseconds)

A value of 0 disables the timer (a zero-length silence period).

DEFAULT: 1500 (1.5 seconds)

You can control the mediatype and encoding of recognition results using the following parameters:

Parameter

Description

Value

server.mrcp2.osrspeechrecog.mrcpdefaults.VSP.server.
osrspeechrecog.result.mediatype

Specifies the media type of the recognition result returned to the application.

MIME media type supported.

DEFAULT: application/x-vnd.speechworks.emma+xml;strictconfidencelevel=1;mrcpv=2.06

server.mrcp1.osrspeechrecog.nlsml.encoding

Inserts an XML header before the NLSML results.

Character encoding type

DEFAULT: ISO-8859-1

server.mrcp1.osrspeechrecog.result.mediatype

Specifies the media type of the recognition result returned to the application.

Media type

DEFAULT: application/x-vnd.speechworks.emma+xml;strictconfidencelevel=1

Configuring text-to-speech resources

This section describes the configuration of text-to-speech resources on Speech Server using Management Station.

Set these parameters if your voice browser uses MRCPv2. (If your browser uses MRCPv1, it ignores these parameters.) You can control several types of activity:

Parameter Description Value

Audio processing

server.mrcp2.rsspeechsynth.audioThreadNumber

Specifies the number of sending threads.

Integer

DEFAULT: 20

server.mrcp2.rsspeechsynth.errorOnNoAudio

Determines whether it is an error when Vocalizer does not generate audio.

Boolean

DEFAULT: 0

Speed of RTP stream

server.mrcp2.rsspeechsynth.rtpSendRate

Specifies the number of audio samples sent per second.

Integer: 1–INT_MAX samples.

DEFAULT: 8000

server.mrcp2.rsspeechsynth.rtpBufferFillMultiplier

Adjusts the RTP sending speed by filling packets more quickly.

Integer: 1–10.

DEFAULT: 2

server.mrcp2.rsspeechsynth.rtpLowerBoundarySamples

Adjusts the RTP sending speed by setting the minimum number of samples that can be sent ahead.

Integer: 0–rtpUpperBoundarySamples samples.

DEFAULT: 300

server.mrcp2.rsspeechsynth.rtpUpperBoundarySamples

Adjusts the RTP sending speed by setting the maximum number of samples that can be sent ahead.

Integer: rtpLowerBoundarySamples–INT_MAX samples.

DEFAULT: 600

Enable/disable use of cookies with Internet fetches

server.mrcp2.rsspeechsynth.cookie.enable

Enables the use of cookies for retrieving files.

Boolean

DEFAULT: 0

Vocalizer input/output

server.mrcp2.rsspeechsynth.plainTextSSMLEncoding

Specifies the default encoding for text/plain MRCP messages.

ISO-8859-1, UTF-8, UTF-16

DEFAULT: ISO-8859-1

server.mrcp2.rsspeechsynth.playsilence

Enables writing silence before and after a prompt.

Boolean

DEFAULT: 0 (no writing silence)

Set these parameters if your voice browser uses MRCPv1. (If your browser uses MRCPv2, it ignores them.) You can control several types of activity:

Parameter

Description

Value
Configuring Vocalizer

server.mrcp1.rsspeechsynth.initialNumber

Specifies the number of synthesizer plug-in instances created during system initialization.

Integer: 0–number of available TTS licenses.

DEFAULT: 0

Audio processing

server.mrcp1.rsspeechsynth.audioThreadNumber

Specifies the number of sending threads.

Integer

DEFAULT: 8

Speed of RTP stream

server.mrcp1.rsspeechsynth.rtpSendRate

Specifies the number of audio samples sent per second.

Integer: 1–INT_MAX samples.

DEFAULT: 8000

server.mrcp1.rsspeechsynth.rtpPacketSamples

Specifies the size of the RTP packet in samples.

Integer: 1–1000 samples. Typical settings are 160 or 240.

DEFAULT: 160

server.mrcp1.rsspeechsynth.rtpBufferFillMultiplier

Adjusts the RTP sending speed by filling packets more quickly.

Integer: 1–10.

DEFAULT: 2

server.mrcp1.rsspeechsynth.rtpLowerBoundarySamples

Adjusts the RTP sending speed by setting the minimum number of samples that can be sent ahead.

Integer: 0–rtpUpperBoundarySamples.

DEFAULT: 300

server.mrcp1.rsspeechsynth.rtpUpperBoundarySamples

Adjusts the RTP sending speed by setting the maximum number of samples that can be sent ahead.

Integer: rtpLowerBoundarySamples–INT_MAX samples.

DEFAULT: 600

Enable/disable use of cookies with Internet fetches

server.mrcp1.rsspeechsynth.cookie.enable

Enables the use of cookies.

Boolean

DEFAULT: 0

Vocalizer input/output

server.mrcp1.rsspeechsynth.plainTextSSMLEncoding

Encoding to plain text for MRCP messages.

ISO-8859-1, UTF-8, UTF-16

DEFAULT: ISO-8859-1