Aim of the interdisciplinary work in the "Acoustic Phonetics" group is the development of recognition and labelling algorithms on the basis of neural nets, taking into account the phonetically important spectral dynamics of the speech signal. The performance of automatic speech recognition is compared with human perception performance for signal sections of equal length. Two aspects of signal dynamics will be given special attention: a) place of articulation, b) word and sentence accent. Both aspects require a processing technique which relates different parts of the signal in the time domain to each other. These aspects have been given little attention in application-oriented work on speech recognition so far.