In this article we describe an application of neural-fuzzy networks to phonetic features recognition over the telephone. We first give a short description of our neural-fuzzy system and talk about knowledge included in it. The advantages of such a combination and its use for helping deaf and hard of hearing people in lip reading are then shortly described. We then present the cases processed for the recognition of the following phonetic features : consonant/vowel, voiced/unvoiced, nasality.