An approach to obtain weighted graphs of words based on phoneme detection

In this paper, we present an approach for phoneme detection and phonetic classification that can be used as a basis for different speech processes, such as phoneme boundary detection, acoustic-phonetic decoding or word-graph construction with acoustic confidence scores. The phonetic classifier that has been developed is based on a phase of acoustic vector clustering in the space of acoustic characteristics, and on a second phase for the association of the acoustic classes with the phonetic units by means of conditional probabilities. We also present methods to build graphs of linguistic units (phonemes, words or semantic units). Phoneme graphs can be applied to acoustic-phonetic decoding tasks or as a preceding step to obtain word graphs. Word graphs can be used in recognition tasks, or they can be converted into graphs of semantic units to be used for understanding tasks. Some recognition experiments and understanding experiments are also presented for a restricted-semantic task about geographical queries.