Generation of a 1000-word vocabulary for a pulse-excited vocoder operating as an audio response unit

The IBM 7772 Audio Response Unit provides standard digital computers with audio-response capability. Vocabulary words and/or sentences are stored in the random-access digital memories of the computer. The coded voice requires 2400 bits per second, and is reconverted into audio by the 7772 audio response unit, which is the synthesizer part of a pulse-excited channel vocoder. It will be shown that the same unit may contain several synthesizers, and thus may answer to several inquiries at the same time, thanks to a multiplexing of the coded voice in the central processing unit. The units are delivered with a 1000-word basic vocabulary in the American English language. To ensure a sufficient level of quality to the synthetic voice, special care must be paid to the vocabulary preparation. The paper will present solutions to problems encountered during vocabulary preparation, especially concerning speaker selection, recordings of words, digitization, quality improvement by software techniques, and message composition. This vocoder is being used with languages other than American English, and with male and female voices. Based on this experience, a comparison will be made between the various languages that have already been processed.