KRrslt—Krypton recognition result

This event is logged when a final result is received from the Krypton recognition engine.

A single result is returned (no partial result events).

In addition to the Tokens used for every event, this event has the following tokens:

Token Meaning
BORT Clock time in milliseconds from first audio packet to start of recognition (based on the endpointer detecting start of speech).
DLMOOV Out of vocabulary (OOV) words introduced by domain language model.
DURS

Amount of speech processed by the Krypton engine in milliseconds.

The value omits any leading samples considered to be silence; therefore, this value will routinely be smaller than EOSS (which includes the leading silence samples).

EFFBUILTINS

"Effective" builtin grammar weights, or those set by the engine during recognition (which may not be the same as those specified by the application—see BUILTINS token in KRrcst event—due to allocation rules).

Specifically, a map of builtin:weight values used in the Recognize command. Weight values are integers.
EFFWEIGHTS

"Effective" grammar (domain language model) weights. (May not be the same as those specified by the application—see WEIGHTS token in KRrcst event—due to allocation rules.)

Specifically, a map of the ID:weight values used in the Recognize command. Weight values are integers.
EFFWORDSETS "Effective" wordset weights. Array of objects identifying the weight associated with a subset of wordsets. Weight values are integers.
EORT Clock time in milliseconds from first audio packet to recognition result available. Measured in real time from the arrival of the first packet of the input stream.
EOSS

End-of-speech signal: where in the input stream the endpointer declared "end of speech". Expressed in milliseconds.

This value includes any leading samples considered to be silence.
KRCONFx Confidence value for n-best item, where x is an integer enumerating each n-best result returned (see Note below). Values can range from 0 to 999.
MEDIA Media type of the audio stream. For example: audio/L16;rate=8000
NBST Number of n-best items. Not emitted when no n-best item returned.
RCPU

Total CPU milliseconds (user and system) consumed during recognition. Note that this value covers CPU from the start of the Recognize command up to the point of generating the result.

Note: CPU consumed in operations that unwind resources and apply adaptation (if any) is included in KRrcnd.
RMFFx Recognition text. Minimally formatted result (transcription).
RSTT See Return codes.
RTOKx Encrypted MREC tokens for each n-best result returned.
RTXTx

Recognition text for each n-best result returned.

Formatted text result (transcription) or the output from the Krypton engine's internal text-processing pipeline (Inverse Text Normalization or ITN).
WORDSETOOV Out of vocabulary (OOV) words introduced by any wordsets.

Note: The KRCONFx, RMFFx, RTOKx, and RTXTx token sequence will appear once for each n-best result returned. KRCONF will be empty if the associated confidence property is unavailable.

The Krypton engine logs these return codes (RSTT):

Return code Status
SUCCESS

Recognition was successful. An n-best result was generated and not rejected.

Note: Only results with this return code will be sent for interpretation.
NO-MATCH There was no recognition match (no hypothesis). An n-best result may or may not be generated; that is, it may be empty or was rejected.
NO-INPUT-TIMEOUT Krypton engine (MREC) did not report start of speech and the limit on pre-speech audio was exceeded (no result returned).
RECOGNITION-TIMEOUT Krypton engine (MREC) did not report end of speech and the limit on the amount of audio to process was exceeded (no result returned).
END-OF-STREAM The Krypton engine (MREC) detected start of speech and the NLP service stopped sending audio before Krypton declared end of speech (no result returned)
ERROR A system error occurred that terminated the recognition (no result returned).