Word Reading Device: Design Considerations

One way to obtain a spoken version of printed text is to fabricate connected speech from voice recordings of single words stored in a random-access memory and assembled under control of the text. The construction of such a word reading device (WORD) is within reach of current technology in the electronic data processing field; however, feasibility depends also on the quality and intelligibility of the output speech. An experimental approach to this “language problem” is the subject of a companion paper; the present discussion deals with over-all design problems, some of the interrelationships between language and machine constraints, and an experimental WORD for evaluation studies and use by the blind. The voice library of the experimental device employs tape recordings of word addresses (digital) and spoken words (analog), arranged in pairs. This method provides compact storage, but it may be desirable, in a full-scale WORD, to use an all-digital memory. This appears to be feasible if the voice entries am stored as instructions for synthesis. Thus, certain parallels can be drawn with a type of reading device that will generate speech by the use of rules for synthesis operating on subphonemic units [A.M. Liberman et at., J. Acoust. Soc. Am. 31, 1490–1499 (1959)]. (Research supported by the Prosthetic and Sensory Aids Service, Veterans Administration.)