Machine-to-man communication by speech part 1: generation of segmental phonemes from text

For many years man has been receiving messages from machines in printed form. Teletypes, computer console typewriters, high-speed printers and, more recently, character display oscilloscopes have become familiar in the role that they play in machine-to-man communication. Since most computers are now capable of receiving instructions from remote locations through ordinary telephone lines, it is natural that we ask whether with all of the sophistication that we have acquired in computer usage, we can communicate with the computer in normal speech. On the input of the computer, there is the automatic speech recognition problem, and at the output, the problem of speech synthesis from messages in text form. The problem of automatic speech recognition is substantially more difficult than the speech synthesis problem. While an automatic speech recognizer capable of recognizing connected speech from many individual speakers with essentially no restriction on the vocabulary is many years away, the generation of connected speech from text with similar restrictions on vocabulary is now well within our reach.