Evaluator

Calculates word error rates and handles conversion of CTC predictions to numeric tokens.

Evaluator.sequenceErrorRate(target, prediction)

Calculates the word error rates based on the target and the predicted inputs.

target and prediction are a table of words such as {"example", "of", "words"}.

Evaluator.predict2tokens(predictions, mapper)

Converts predictions of the neural network into a sequence of tokens (characters) via a mapper.

predictions is a tensor of sequence likelihood vectors given by the neural network.

mapper Defines how numeric values are converted to readable tokens (see Mapper).