Dynamic sharings of gaussian densities using phonetic features