This paper presents a parallel processing principle implemented in a network of processing units, which requires a specific spectral representation of speech. This so-called topographic representation can be obtained through non-linear transformations that are described. Robustness to time variations of the resulting speech recognition system is shown to result from the duration of the internal signals that are synchronized and summed by the processing units.
Bibliographic reference. Beroule, Dominique G. (1989): "Management of time distortions through rough coincidence detection", In EUROSPEECH-1989, 1454-1457.