An acoustic profile of speech efficiency

Speaking is generally considered efficient, in that less effort is spent articulating more redundant items. Two possible mechanisms for this optimization are tested. The use of prosodic structure, i.e., lexical stress and sentence accent, to (de-)emphasize (un-)important words, and the facilitation of syllable-articulation by retrieving often used motor-programs from memory. Such a “stored versus computed” principle in syllable articulation would implicitly result in efficient speech because of the correlations between syllableand wordfrequencies. These mechanisms are tested for Dutch speech by means of a hand labeled single-speaker corpus of spontaneous and matched read speech, and an automatically labeled multispeaker corpus of read telephone speech. It is concluded that the use of lexical stress and sentence accent/prominence cannot explain all of the frequency-of-occurrence effects found in speech. Furthermore, at least in unstressed syllables, syllablefrequency effects proved to be more important than wordfrequency effects, leaving room for an articulatory “stored versus computed” mechanism in the optimization of speaking effort.