Measuring the Confusability of Pronunciations in Speech Recognition
Résumé
In this work, we define a measure aimed at assessing how well a pronunciation model will function when used as a component of a speech recognition system. This measure, {\\it pronunciation entropy}, fuses information from both the pronunciation model and the language model. We show how to compute this by effectively composing the output of a phoneme recognizer with a pronunciation dictionary and a language model, and investigate its role as predictor of pronunciation model performance. We present results of this measure for different dictionaries with and without pronunciation variants and counts