An audio response method for CAI services

This paper describes an audio response method, where audio messages are stored in a movable-head magnetic disk memory in the form of speech feature parameters through the PARCOR speech analysis. Desired message data is asynchronously read out of the disk memory according to the readout demand queue into a buffer memory to drive a speech synthesizer. The achievable number of output channels is related with the average message duration and the disk memory characteristics. This method realizes an audio filing unit for maximally 128 output channels, of 0.1-second response time, and with 5000 messages of 15-second duration per 100 MB disk. It is suitable to interactive CAI services for individually paced training.