Improvements for interrogative prosody modeling are proposed.
Hybrid rule-based methods are used in the HMM framework.
Performance is evaluated by expert evaluation and listening tests.
Quality of interrogative prosody is enhanced significantly over baseline.