European Conference on Speech Technology
Edinburgh, Scotland, UK
Altmann (1986) investigates the relative usefulness of different parts of the speech signal for word recognition, and claims that a front end which is able to provide "fine class" (phonemic) information about segments in stressed syllables and "mid class" information about those in unstressed would on average be less helpful for word identification than one which was able to attach a fine class label to the same proportion of segments, but scattered at random throughout words. Altmann concludes that his results "demonstrate the desirability of a front-end which outputs occasional fine class information not systematically, but at random".
We show that Altmann's results are based on a faulty experimental procedure, which leads him to incorrect conclusions; fine class information about segments in stressed syllables is actually slightly more useful than that about randomly chosen segments.
Bibliographic reference. Carter, David / Boguraev, Bran / Briscoe, Ted (1987): "Lexical stress and phonetic information: which segments are most informative?", In ECST-1987, 1235-1238.