This section describes how to score speech-to-text results against a truth transcript.
If a truth transcript corresponding to an audio recording is available, you can use the Scorer
standard task to calculate how accurate the speech-to-text output is. To get an accurate estimate, the transcript must be verbatim–that is, every word must be transcribed, regardless of whether it is grammatically correct. The key metrics that the scorer reports are general word precision, recall, and the F-measure.
|