This event logs detailed statistics of recognizer processing after every recognition event. This event is not logged by default because its primary purpose is for troubleshooting by Nuance technical support.
The typical use of this event is when you’re experiencing performance problems, and a technical support engineer requests you enable this event for a period of time and reply with the resulting call logs.
To enable, use swirec_save_comp_stats.
The key/value pairs describe audio during, CPU usage, grammar counts, and memory usage during processing by the recognizer.
In addition to the Tokens used for every event, this event has the following tokens:
|
Token |
Meaning |
|---|---|
|
AUDDUR |
The duration of audio processed by the recognizer in milliseconds. |
|
AUDFRAMES |
The duration of audio processed by the recognizer in frames. |
|
MSTOTAL |
Total CPU used for recognition in milliseconds. |
|
MSFASTMATCH |
CPU used for the fastmatch of internal mrec processing in milliseconds. |
|
MSMFCC |
CPU used for feature computation in milliseconds. |
|
MSFORWARD |
CPU used for first pass recognition in milliseconds. |
|
MSBACKWARD |
CPU used for generating the lattice, n-best generation, and second-pass recognition in milliseconds. |
|
MSRESCORE |
CPU used for final n-best scoring of the final pass. The final pass could be the first- or second-pass recognition). Includes confidence calculations and some natural language resorting in milliseconds. |
|
MSPARSE |
CPU used to parse semantics on the n-best entries in milliseconds. |
|
MSNBEST |
|
|
MS2NDPASSSETUP |
CPU used for initializing the second pass in milliseconds. |
|
MS2NDPASSSCORE |
CPU used for second pass rescoring in milliseconds. |
|
MSADAPTATION |
CPU used to adapt models in milliseconds. |
|
NCONTEXT |
The number of parallel grammars in the recognition. |
|
NWORDS |
The maximum number of words used in a grammar. |
|
NDISTCOMP |
Total number of distance computations for the whole recognition (first pass, second pass, and so on). |
|
NARCACTIVE |
Total number of active model arcs (HMMs). |
|
NSTATEACTIVE |
Total number of active HMM states. |
|
NSENTTRIED |
Number of sentences extracted from decoders to build the semantic n-best list. |
|
NNBEST |
Number of entries in the n-best list. |
|
NFSMARC |
Number of non-epsilon FSM arcs queried from the constraint graph. |
|
MEMFSM |
Amount of memory used for static grammar networks in bytes. |
|
MEMUTT |
Amount of memory dynamically allocated at runtime during the recognition in bytes. |
