ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

4aPP9. Probability theory and speech perception.

Arthur Boothroyd

City Univ. of New York, 33 W. 42 St., New York, NY 10036

Fletcher recognized the value of probability theory in developing a quantitative approach to speech perception---hence the articulation index (AI). He also recognized the need to allow for violations of the independence assumptions that underly basic probability theory. For example, in nonsense syllables, the probability (p[sub w]) of recognizing a word equals p[sub p][sup n], where p[sub p] is the probability of recognizing the constituent phonemes and n is the number of phonemes per word. In real words, however, p[sub w]=p[sub p][sup j], where n(greater than or equal to)j(greater than or equal to)1. Similarly, extending the methods of AI, the probability of recognition of words in sentences ([sup s]p[sub w]) can be shown to be related to the probability of recognition in isolation ([sup i]p[sub w]) by the equation [sup s]p[sub w]=1-(1-[sup i]p[sub w])[sup k], where k is an exponent reflecting the contribution of the sentence context. From these two basic equations one can derive relationships among many measures of speech perception, ranging from the phonetic level to the sentence level. The empirical values of the exponents j and k can be used both to quantify the effects of various structural and contextual constraints and to assess an individual's use of those constraints. [Work supported by NIDCD Grant No. 10078.]