Modeling Speech Melody as Communicative Functions with PENTAtrainer2

This paper presents PENTAtrainer2, a semi-automatic software package written as Praat plug-in integrated with Java programs, and its applications for analysis and synthesis of speech melody as communicative functions. Its core concepts are based on the Parallel Encoding and Target Approximation (PENTA) framework, the quantitative Target Approximation (qTA) model, and the simulated annealing optimization. This integration allows it to globally optimize for underlying pitch targets of specified communicative functions. PENTAtrainer2 consists of three computational tools: Annotation tool for defining communicative functions as parallel layers, Learning tool for globally optimizing pitch target parameters, and Synthesis tool for generating speech melody according to the learned pitch targets. Being both theory-based and trainable, PENTAtrainer2 can serve as an effective tool for basic research in speech prosody.