VoiceXML 2.1 Development Guide

| TOC

Advanced Nuance ASR Property settings

If you want to have an extra degree of fine-tuning to your application properties, then you will need to look no further. The listing that follows details the internal Nuance-specific properties that you can utilize to further refine your application. However, be advised, that these properties must be used carefully and sparingly. Setting any of these properties can conceivably prove detrimental to recognition accuracy, and might even throw an error within the application, depending on how liberally you place these in the xml code. Therefore, it is strongly advised that you be extraordinarily careful when using any of these properties.

Note: Any 'core' Nuance property used on the Prophecy platform must be prefixed by 'com'; to illustrate:

<property name="com.nuance.client.AllowBargeIn" value="true"/>

Recclient Parameters

Property	Value
com.nuance.client.AllowBargeIn	`true\|false`

Details: This property controls whether or not a caller is allowed to barge in on a TTS or audio prompt. Essentially the same as specifying <prompt bargein>

Property	Value
com.nuance.client.HotWordMaxSeconds	`0.00 +`

Details: This property controls the maximum duration (in seconds) that an utterance can be to be evaluated against the hot word. This parameter, along with client.HotWordMinSeconds, can be used to tune performance by preventing the recognizer from evaluating utterances that are too short or too long to be the intended hotword. This property defaults to a value of 1.70 seconds.

Property	Value
com.nuance.client.HotWordMinSeconds	`0.00 +`

Details: This property controls the minimum duration (in seconds) that an utterance can be to be evaluated against the hot word. This parameter, along with client.HotWordMinSeconds, can be used to tune performance by preventing the recognizer from evaluating utterances that are too short or too long to be the intended hotword. This property defaults to a value of 0.30 seconds.

Property	Value
com.nuance.client.MinPreSpeechSilenceSecs	`0.00 +`

Details: The MinPreSpeechSilenceSeconds property allows the developer to govern how much silence, (in seconds), is required after recognition begins, (i.e., when the interpreter is expecting a response from the caller). If the user speaks before the required minimum number of pre-speech silent seconds, the timeout behavior generates a fake final result with the enumerated type value RecResultType::SPEECH_TOO_EARLY, (which is a catchable event type), and will then reprompt the caller. This fake result is delayed until end-of-speech occurs. The value for this property defaults to 0.10; to allow the caller to speak at any point after recognition begins, you can set this property to 0.0.

Property	Value
com.nuance.client.NoSpeechTimeoutSecs	`0.00 +`

Details: The above property is used for determining the amount of silence, (in seconds), to allow after recognition begins or, if barge-in is enabled, after playback completes. If the user does not speak within the specified number of seconds, interpreter will throw a NO_SPEECH_TIMEOUT event, (longhand for the noinput event).. If you set this parameter to 0, the timeout is disabled. The default value of 32000 seconds also effectively means that the timeout never occurs.

Property	Value
com.nuance.client.RecognizerTooSlowTimeoutSecs	`0.00 - 60.00`

Details: Specifies the maximum number of seconds to wait for the final recognition result from the recognizer after end-of-speech In all likelihood, you will not have to tweak this setting at all. If a final result does not arrive within the specified number of seconds after end-of-speech, the timeout behavior generates a fake final result with the enumerated type value RecResultType::RECOGNIZER_TOO_SLOW_TIMEOUT. Set this parameter to 60.00, (default) to prevent the timeout from occurring as often as possible. If you set this parameter to 0, the timeout is disabled.

Property	Value
com.nuance.client.TooMuchSpeechTimeoutSecs	`0.00 - 600.00`

Details: This property specifies the maximum number of seconds of speech to allow after the user begins speaking, allowing you to dictate how long a user's utterance can be. When an utterance length exceeds the value for this property, the event of TOO_MUCH_SPEECH_TIMEOUT will be thrown. Set this parameter to its maximum value of 600, (default), to avoid this timeout from occurring. If you set this parameter to 0, the timeout is disabled.

Property	Value
com.nuance.dtmf.BufferedKeys	`(no value)`

Details:Setting this property to an empty value, <property name="com.nuance.dtmf.BufferedKeys" value=""/> allows the edeveloper to clear the DTMF digit buffer for the form/form item that the caller is currently in. A much easier method is to simply add an empty <prompt bargein="false"></prompt> within the code to clear the digit buffer.

Endpointing properties

Property	Value
com.nuance.ep.AdditionalEndSilence	`0.00 - 5.00`

Details: Specifies the number of seconds of post-speech silence to include with the endpointed speech when it is recorded or recognized. Should you wish to tack on additional silence to the end of a recorded audio file, enable this property within your code from the default value of 0.65. Note that the value specified for this parameter is really a minimum value, not an absolute value.

Property	Value
com.nuance.ep.AdditionalStartSilence	`0.00 - 5.00`

Details: Specifies the number of seconds of pre-speech silence to include with the endpointed speech when it is recorded or recognized. Should you wish to tack on additional silence to the end of a recorded audio file, enable this property within your code from the default value of 0.35.

Property	Value
com.nuance.ep.BargeInMinNoiseFloor	`5 - 20`

Details: This property governs the volume of speech that is required in order to successfully barge in on a prompt. For callers in a noisy environment, you can adjust this property upwards from it's default value of 13, but bear in mind that this may increase missed barge-ins with quiet speakers. Use this property with extreme caution.

Property	Value
com.nuance.ep.EndSeconds	`0.20 - 10.00`

Details: Specifies the minimum amount of silence required to indicate that the end of speech has occurred. Setting this to a higher value than the default of 2.0 would make it less likely that the caller's utterance will be cut off, but will impact the time it takes to return the recognized result. It is advisable that you adjust this property upwards at the field level when you expect a longer utterance from the caller, thus allowing for breaks and pauses in the user's speech.

Property	Value
com.nuance.ep.EndSecondsOnGrammarEnd	`0.10 - 10.00`

Details: Essentially, this property is an internal Nuance ASR property setting that matches up with the W3C 'completetimeout' property. Adjustments from the default value of 0.25 will specify the amount of time, in seconds, that is required to determine the end of a caller's utterance when the last recognized word matches up with the utterance defined in the grammar file itself.

Property	Value
com.nuance.ep.EndSecondsOnGrammarFinal	`0.10 - 10.00`

Details: The setting for this property defines the number of seconds of silence required to determine the end of speech when the last recognized word is an allowable final word for the utterance, as defined in the current recognition grammar. If the word can only appear in the final position, then the setting of ep.EndSecondsOnGrammarEnd is used. The value for this property is equal to 0.75, unless adjusted by the developer.

Property	Value
com.nuance.ep.StartSeconds	`0.06 - 1.00`

Details: This property allows you to adjust the minimum duration of speech required to indicate that the talker has begun to speak, thus allowing recognition to begin. Usually, the developer will never have reason to change this setting from its default setting of 0.30.

Property	Value
com.nuance.ep.ThresholdSnr	`5 - 30`

Details: This is the 'behind the scenes' Nuance property that gets adjusted when the developer tweaks the sensitvity property. Essentially, the value entered for this property dictates the amount of noise required for the recognition engine to detect start-of-speech and end-of-speech. To reduce the likelihood of false detection in noisy environments, you can increase the value of this parameter, but this will make all callers need to speak at a greater volume in order for their utterance to be recognized. Developers should rarely need to modify this setting from the default of 12, and it is advised that you use this property with extraordinary care in your code.

Property	Value
com.nuance.ep.ThresholdSnrEndOffset VWS2.0 only	`-30 - 0`

Details: The value of this parameter is applied to the threshold specified by parameter ep.ThresholdSnr to detect the end-of-speech. A larger negative number makes it more likely that phrases ending weakly are classified as speech.

Recognition Properties

Property	Value
com.nuance.rec.BacktraceFinalsOnly	`true\|false`

Details: This property, when set to TRUE (the default), will dictate that the system return a recognition result onlywhen the complete utterance match is acheived. For example, if a recognition grammar requires 5 digits but the speaker utters only 3 digits, the utterance is rejected if com.nuance.rec.BacktraceFinalsOnly=TRUE. Otherwise, the 3-digit response is returned.

Property	Value
com.nuance.rec.ConfidenceRejectionThreshold	`0 - 100`

Details: This property sets the confidence value that a user utterance must score in order to be considered a valid grammar match. When the value for a utterance is lower than this threshold, the system replaces the recognized result with an empty string. You can set this parameter to 0 to avoid any rejection based on the confidence score of a recognition result. Note that in all likelihood, you will never need to adjust this setting from its default value of 45.

Property	Value
com.nuance.rec.DoNBest	`true\|false`

Details: This property, when set to 'true', (default), will allow more than one recognition result when an ambiguous grammar utterance is encountered by the interpreter, thus allowing for N-best post processing.

Property	Value
com.nuance.rec.FinalResultTimeoutMs	`500 +`

Details: This property sets the timeout for the amount of time the RecServer is allowed to return a final result after processing ends (for example, if the operation is aborted). usually, you will never have the need to adjust this setting from its default value of 60000.

Property	Value
com.nuance.rec.GenConfidence	`true\|false`

Details: This property specifies whether or not to return a confidence score with each grammar match. There is virtually no need at all to ever change this setting from the default value of 'true'.

Property	Value
com.nuance.rec.GenSlotConfidence	`true\|false`

Details: This property denotes whether the system should generate confidence scores for each individual slot filled in the recognition result. By default, this is set to true, which allows you to access the scores for each slot with RecResultConfidenceForIthSlot.

Property	Value
com.nuance.rec.GrammarWeight	`0.00 - 100.00`

Details: Controls the grammar processing weight -- the overall 'validity' scores during recognition. This property can be used in conjunction with rec.Pruning to control speed/accuracy trade-offs.

Property	Value
com.nuance.rec.MaxNumInterpretations	`1 - 500`

Details: This property controls how many results that the interpreter should return upon an ambiguous grammar match. the default setting for this is '10', but if you are using a grammar and Nbest post processing that could conceivably have a greater amount of matches, (such as a company directory), then you will likely want to increase the setting of this property.

Property	Value
com.nuance.rec.NumNBest	`1 - 500`

Details: This setting controls the maximum amount of results that can be returned when performing N-best post processing. Increase the default value of '10' when you need to return more results from an ambiguous match. Do note that increasing this value also increases the time it takes to return results from a grammar match.

Property	Value
com.nuance.rec.PPR	`true\|false`

Details: Specifies whether the recognizer should perform phonetic pruning upon a callers utterances.

Property	Value
com.nuance.rec.PruneRejects	`true\|false`

Details: Specifies whether the recognition process should prune phrases that are likely to be rejected.

Property	Value
com.nuance.rec.Pruning	`400 - 5000`

Details: Controls the trade off between speed and accuracy upon a grammar match. A smaller number increases recognition speed but decreases accuracy, while a larger number increases accuracy but slows performance.

Property	Value
com.nuance.rec.RejectWeight	`-500 - 60`

Details: Specifies the weighting value used to determine when a callers utterance is rejected.

Speech Channel Properties

Property	Value
com.nuance.sc.CallTimeoutSecs	`1 +`

Details: Sets the timeout for speech channel functions that attempt to establish a call connection, including placing a call, transferring a call, or setting up a conference. If the connection cannot be established within this number of seconds, the function returns with a timeout failure status.

Property	Value
com.nuance.sc.DropCallWhenSpeakerHangup	`true\|false`

Details: Controls whether all sides of a tromboned call are dropped if the party being listened to hangs up. The default setting for this property is 'true'.

Property	Value
com.nuance.sc.KillPlaybackOnDtmf	`true\|false`

Details: This setting controls whether or not a caller's DTMF input can interrupt a TTS or audio prompt. Should you wish to allow any prompts to continue without interruption upon DTMF input, change this property's setting to 'false'.

Property	Value
com.nuance.sc.RecordingNoSpeechTimeoutSecs	`0 - 32000`

Details: This property specifies the number of seconds of silence to allow after recording begins, or before a recording operation times out because no speech is detected. If you set this to 0, the timeout is disabled. The default value for this property is set to '20'.

Property	Value
com.nuance.sc.RecordingTooMuchSpeechTimeoutSecs	`0 - 32000`

Details: This setting controls the maximum allowable number of seconds of speech a caller may record. If no end-of-speech is detected when this number of seconds have passed since the caller started speaking, the recording operation times out. If you set this to 0, the timeout will be disabled.

Property	Value
com.nuance.sc.StopRecordingOnDtmf	`true\|false`

Details: This setting specifies whether or not any recording session in progress may be ended by dtmf entry from the caller.

Advanced Motorola Properties

Motorola does provide quite a few property settings as well, although most of these are used for internal settings, and have no place at all in this listing. However, the ones which may serve useful; to developers are listed here.

Property	Value
com.mot.debugflags	`(see below table)`

Details: This property allows the developer to set the debug flags for the <meta maintainer> emailing feature of the the Prophecy IVR platform. Using these allows the developer to have a much finer grain of control in regards to what email alerts get sent to the code maintainer. The potential to fill up an mail inbox with system messages is a very real possibility in the event of an application failure with hundreds of users involved. Consider yourself warned, use this <meta maintainer> feature judiciously.

Setting	Explanation
E	Character encoding related messages
P	Prints the current VXML document on each listen for user input
T	TTS call information
b	Billing Messages
c	Cookie related messages
d	Document and dialog entry/exit messages
e	ECMAScript interpreter messages
f	Fetch messages
h	HTTP response header messages
p	Property change event messages
r	Speech recognition messages
s	Symbol table messages
w	Fetch work queue messages
x	Execution context messages
z	On each fetch, print out the contents of the fetch result
$	Print out performance cost messages, (Parse times, compilation times, fetch times)

Property	Value
com.mot.recognitionbasedbargein	`true\|false`

Details: This setting allows a developer to specify whether 'hotword' bargein behavior is enabled for the entire application. To illustrate, when set to 'true', a bargein will only be allowed & acted upon when the user's utterance matches an active grammar, otherwise, the TTS or audio file in question will continue playing uninterrupted, (thus, nullifying any chances for a 'nomatch' event). When set to 'false', (default), then any utterance may barge in while an utterance is expected, although if the utterance is not a valid grammar match, then a 'nomatch' event will be raised.

ANNOTATIONS: EXISTING POSTS

0 posts - click the button below to add a note to this page

Nuance ASR Property settings

| TOC

| I: GSL Grammars