Desarrollo de un corrector ortográfico para aplicaciones de conversión texto-voz

Telefónica I+D has developed an automatic correction method specially focused on the adaptation of Internet text in order to be read by our text-to-speech system. The correction model detects errors by searching words in a dictionary, and then (if the word isn ́t found) correction candidates are generated by making elemental transformations (such as insertion, deletion, substitution or transposition). The selection of the best candidate takes into account the word probability (its frecuency in the training corpus), the probability of the elemental transformation (confusion matrices), the context (language model) and the text correction level (reliability factor)