Special SWI_ keys

Nuance supports proprietary SWI_ keys that you can set in grammars, or that are set automatically by Recognizer after an utterance recognition.

Note: Nuance Recognizer supports all keys. Dragon Voice supports SWI_literal, SWI_spoken, SWI_meaning, SWI_utteranceSNR.

Use this key to set variables in application URIs, and use ECMAScript in grammars to process them:

• SWI_vars: Passes application variables to grammars.

During recognition, use ECMAScript in grammars to get and set these keys:

• SWI_literal: Shows raw recognized text.
• SWI_literalConfidence: Shows confidence for each slot.
• SWI_spoken: Shows recognized text for n-best entries.

Set these keys in grammars to control n-best processing and confidence scoring:

• SWI_decoy: Improves the rejection of out-of-vocabulary items.
• SWI_disallow: Prevents invalid n-best entries.
• SWI_safeKey: Passes recognition results to log files even if security settings are enabled.
• SWI_scoreDelta: Attaches weights to specific parses.

The recognizer sets these keys and returns them to the application in recognition results:

• SWI_attributes: Lists result attributes.
• SWI_bestModel: Provides a source recognition model.
• SWI_confidence: Shows confidence for each n-best entry—scripts can get and set this key.
• SWI_grammarName: Shows the grammar for each n-best entry.
• SWI_literalTimings: Shows word durations and word confidence scores.
• SWI_meaning: Sets semantic meaning of recognized text—scripts can get and set this key.
• SWI_rawScore: Shows the raw recognition score.
• SWI_semanticSource: Shows how meaning was assigned.
• SWI_ssmConfidences: Shows confidence scores for results in a statistical semantic model.
• SWI_ssmMeanings: Shows semantic meanings.
• SWI_utteranceSNR: Shows signal-to-noise ration for n-best entries.