Letter-to-sound pronunciation prediction using conditional random fields