|
KeenASR Framework v2.2 (c662ac9)
Keen Research
|
Per-call configuration for a text-alignment operation. More...
#include <KIOSTextAligner.h>
Properties | |
| float | insertCost |
| Cost of inserting a recognized token (one not present in the reference). | |
| float | deleteCost |
| Cost of deleting a reference token (one not present in the recognized text). | |
| float | substituteCost |
| Cost of substituting one token for another. | |
| BOOL | detectRepetitions |
| When YES, the result populates repetitionRefIndices for words the reader appears to have stuttered or repeated. | |
| BOOL | filterNoiseTokens |
| When YES, recognized tokens whose text begins with '<' (e.g. <SPOKEN_NOISE>, <UNK>) are dropped before alignment. | |
Per-call configuration for a text-alignment operation.
Defaults reproduce classic Levenshtein word edit distance with unit costs and noise-token filtering enabled. For most oral-reading and WER use cases the defaults are appropriate.
|
readwritenonatomicassign |
Cost of deleting a reference token (one not present in the recognized text).
Default: 1.0.
|
readwritenonatomicassign |
When YES, the result populates repetitionRefIndices for words the reader appears to have stuttered or repeated.
A repetition is reported when two adjacent recognized tokens are identical but only one of the pair was matched against the reference. Default: NO.
|
readwritenonatomicassign |
When YES, recognized tokens whose text begins with '<' (e.g. <SPOKEN_NOISE>, <UNK>) are dropped before alignment.
Leave on when the recognized text comes from an ASR result; turn off when aligning plain reference vs. plain hypothesis. Default: YES.
|
readwritenonatomicassign |
Cost of inserting a recognized token (one not present in the reference).
Default: 1.0.
|
readwritenonatomicassign |
Cost of substituting one token for another.
When substituteCost > insertCost + deleteCost the aligner degenerates to LCS-style behavior (no substitutions emitted). Default: 1.0.