On a hybrid time domain-LPC technique for prosody superimposing used for speech synthesis