Skip to main content

Table 5 Percentages of correctly and incorrectly classified decoding results of the two different subsets and the total set using the global EER threshold and all predictors. (a) Percentages of decoding result classification on the set where the correct transcription was in the language model. (b) Percentages of decoding result classification on the set where the correct transcription was not present in the language model. (c) Percentages of decoding result classification on the whole dataset.

From: Optimizing Automatic Speech Recognition for Low-Proficient Non-Native Speakers

(a)
   Actual
   Correct Incorrect
Predicted Correct 80.8% 3.0%
  Incorrect 9.2% 7.0%
(b)
   Actual
   Correct Incorrect
Predicted Correct 8.3%
  Incorrect 91.7%
(c)
   Actual
   Correct Incorrect
Predicted Correct 40.4% 5.6%
  Incorrect 4.6% 49.4%