Tempo Control in Speech Synthesis by Prosodic Phrasing

Tempo control in most speech synthesisers is performed by linear time-scaling although tempo change in human speech shows a non-linear nature. In a perception experiment with a German speech synthesiser it was found that the versions with adjusted prosodic breaks and pauses are preferred over the linear versions for two fast rates and particularly for "very slow". However, the model for "rather slow" needs a refined syntax-prosody mapping.