Computer Simulation of Spontaneous Speech Production

COMPUTER SIMULATION OF SPONTANEOUS SPEECH PRODUCTION Bengt Sigurd Dept of Linguistics and Phonetics Helgonabacken 12, S-223 62 Lund, SWEDEN ABSTRACT This paper pinpoints some of the problems faced when a computer text production model (COMMENTATOR) is to produce spontaneous speech, in particular the problem of chunking the utterances in order to get natural prosodic units. The paper proposes a buffer model which allows the accumula- tion and delay of phonetic material until a chunk of the desired size has been built up. Several phonetic studies have suggested a similar tempo- rary storage in order to explain intonation slopes, rythmical patterns, speech errors and speech dis- orders. Small-scale simulations of the whole ver- balization process from perception and thought to sounds, hesitation behaviour, pausing, speech errors, sound changes and speech disorders are pre- sented. 1. Introduction Several text production models implement- ed on computers are able to print grammatical sen- tences and coherent text (see e.g. contributions in All~n, 1983, Mann & Matthiessen, 1982). There is, however, to my knowledge no such verbal production system with spoken output, simulating spontaneous speech, except the experimental version of Commentator to be described. The task to design a speech production system cannot be solved just by attaching a speech synthesis device to the output instead of a printer. The whole production model has to be reconsidered if the system is to produce natural sound and pro- sody, in particular if the system is to have some psychological reality by simulating the hesitation pauses, and speech errors so common in spontaneous speech. This paper discusses some of the prob- lems in the light of the computer model of verbal production presented £n Sigurd (1982), Fornell (1983). For experimental purposes a simple speech synthesis device (VOTRAX) has been used. The Problem of producing naturally sounding utterances is also met in text-to-speech systems (see e.g. Carlson & Granstr~m, 1978). Such systems, however, take printed text as input and turn it into a phonetic representation, eventually sound. Because of the differences between spelling and sound such systems have to face special prob- lems, e.g. to derive single sounds from the letter combinations t__hh, ng, sh, ch in such words as the, thing, shy, change. 2. Co,~entator as a speech production system The general outline of Con~entator is presented in fig. I. The input to this model is perceptual data or equivalent values, e.g. infor- mation about persons and objects on a screen. These primary perceptual facts constitute the basis for various calculations in order to derive secondary facts and draw concluslons about movements and re- lations such as distances, directions, right/left, over/under, front/back, closeness, goals and in- tentions of the persons involved etc. The Commentator produces comments consisting of gram- matical sentences making up coherent and well- formed text (although often soon boring). Some typical comments on a marine scene are: THE SUB- 79 MARINE IS TO THE SOUTH OF THE PORT. IT IS APPROACH- ING THE PORT, BUT IT IS NOT CLOSE TO IT. THE DESTROYER IS …

Content truncated to 500 words