Localización de Palabras basada en Grafos de Fonemas

In this paper we propose the use of phoneme graphs for word spotting tasks. The proposed phoneme graphs are built using phonetic probabilities estimated at frame level. We also propose an error model on phoneme graphs which allows to simplify the exploration algorithms used for finding phonetic sequences. A phoneme duration model is also applied for avoiding the detection of too short phonetic sequences, which helps to reduce the number of false positive detections.