Parallelism and the transputer in the automatic translation of text to speech

The automatic translation of text to speech is most useful when not restricted to a prestored vocabulary. The potential quality of speech output has improved greatly as the complexity of speech synthesis chips has increased. This improvement in quality necessitates greater computing resources for handling such processes as text normalization and the application of an allophone string generation algorithm. Many of the processes involved can be run concurrently. A multitransputer system programmed in Occam has been constructed to perform automatic text-to-speech conversion. From the timings obtained for a single-transputer system it can be seen that 86% of the time the transputer system is idle, waiting for the speech synthesis section to finish outputting allophones. In the case of the multitransputer system, this idle time is increased, as expected, to 89%. The speech quality is good although rather robotic due to the fixed accessing time of the speech synthesis chip and its lack of inflection. It is concluded that a transputer system is ideally suited to this type of application