Evaluating and Rewarding LALMs for Expressive Role-Play TTS via Mean Continuation Log-Probability

Recent advances in Large Audio Language Models (LALMs) have propelled Text-to-Speech (TTS) toward interactive role-play scenarios, necessitating high expressiveness and strict adherence to role-play instructions. However, existing models struggle to maintain stylistic consistency with character profiles and scene descriptions across multi-turn dialogues. A critical bottleneck remains the lack of objective metrics for quantifying speaking style. To bridge this gap, we propose Mean Continuation Log-Probability (MCLP) as both an evaluation metric and a reward signal, validated on LALM-based RP-TTS tasks. Critically, we leverage the In-Context Learning capability of pre-trained LALMs to design MCLP via a reverse-continuation paradigm. This metric quantifies stylistic consistency by measuring the likelihood of the ground-truth speech conditioned on the generated speech. Furthermore, we employ MCLP as a reinforcement learning reward, which significantly enhances the style alignment between the generated speech and the scene/character instructions. To facilitate evaluation, we construct an RP-TTS dataset featuring rich scene and character annotations. Experimental results demonstrate that our method significantly outperforms strong LALM baselines across both objective and subjective metrics.