A nucleus-based timing model applied to multi-dialect speech synthesis by rule