edu.cmu.sphinx.result
Interface ConfidenceScorer
- All Superinterfaces:
- Configurable
- All Known Implementing Classes:
- AbstractSausageMaker, MAPConfidenceScorer, PivotSausageMaker, SausageMaker
public interface ConfidenceScorer
- extends Configurable
Computes confidences for a Result. Typically, one is interested in the confidence of the best path of a result, as
well as the confidence of each word in the best path of a result. To obtain this information, one should do the
following:
ConfidenceScorer scorer = (ConfidenceScorer) ... // obtain scorer from configuration manager
Result result = recognizer.recognize();
ConfidenceResult confidenceResult = scorer.score(result);
// confidence for best path
Path bestPath = confidenceResult.getBestHypothesis();
double pathConfidence = bestPath.getConfidence();
// confidence for each word in best path
WordResult[] words = bestPath.getWords();
for (int i = 0; i < words.length; i++) {
WordResult wordResult = (WordResult) words[i];
double wordConfidence = wordResult.getConfidence();
}
Note that different ConfidenceScorers have different definitions for the 'best path', and therefore their
getBestHypothesis
methods will return different things. The MAPConfidenceScorer
returns the highest scoring path. On the other hand, the SausageMaker
returns the path where all the words have the highest confidence in their
corresponding time slot.
Method Summary |
ConfidenceResult |
score(Result result)
Computes confidences for a Result and returns a ConfidenceResult, a compact representation of all the hypothesis
contained in the result together with their per-word and per-path confidences. |
score
ConfidenceResult score(Result result)
- Computes confidences for a Result and returns a ConfidenceResult, a compact representation of all the hypothesis
contained in the result together with their per-word and per-path confidences.
- Parameters:
result
- the result to compute confidences for
- Returns:
- a confidence result