Specifies the minimum confidence score. Nuance Recognizer rejects utterances with scores below this value.
|
Value |
DEFAULT: 0 (all utterances accepted) |
|
How to set |
Shown in precedence order (see Rules of parameter precedence): In a VoiceXML application, using a <property> element. In the browser, using the Confidence-Threshold header in a RECOGNIZE or SET-PARAMS method. In a parameter grammar. On the Speech Server. In Management Station set on the Nuance recognition service. If not using Management Station, set in a Recognizer configuration file (User-nrsxx.xml). |
|
Usage |
Frequently used by application developers during development and for tuning. Most often it is set as a VoiceXML property to tune individual collections. Applicable for Nuance Recognizer. Not applicable for Dragon Voice recognition. |
The confidence level is a measure of Recognizer’s confidence that it has matched the correct result (assuming that the caller said something in Recognizer’s vocabulary). Recognizer computes scores in the range from 0 to 1000, with 1000 being the highest confidence, and then compares the result to the confidencelevel parameter. (The call logs report the confidencelevel in the same 0 to 1000 range.) If Recognizer’s score is below the parameter value, then Recognizer returns a SWIrec_STATUS_NO_MATCH status.
Applications use confidence scores to decide whether to confirm what callers have said. For example, a score above 750 might be accepted, a score between 200–750 might require confirmation, and a score below 200 might require rejection. This parameter sets the minimum threshold for a recognition match, and allows rejection decisions inside Recognizer (instead of forcing the application to evaluate and reject).
The default is 0, which always sets the result status to success providing that Recognizer was able to populate the n-best list. (This default means that the application must make all decisions for acceptance, confirmation, and rejection.)
A typical application sets this parameter via a VoiceXML property or MRCP header. Note that these methods use different ranges of values, rather than the expected range of 0 to 1000. Speech Server automatically re-scales these to the range used by Recognizer.
The example below sets a value of 0.2:
<property name="confidencelevel" value="0.2"/>
The browser must convert this to an MRCP header. The value is different for MRCP v1 and v2:
Confidence-Threshold: 20
Confidence-Threshold: 0.2
