Emphasized Accent Phrase Prediction from Text for Advertisement Text-To-Speech Synthesis

Realizing expressive text-to-speech synthesis needs both text processing and the rendering of natural expressive speech. This paper focuses on the former as a front-end task in the production of synthetic speech, and investigates a novel method for predicting emphasized accent phrases from advertisement text information. For this purpose, we examine features that can be accurately extracted by text processing based on current Text-tospeech synthesis technologies. Among features, the word surface string of the main content and function words and the part-of-speech of main function words in an accent phrase are found to have higher potential on predicting whether the accent phrase should be emphasized or not through the calculation of mutual information between emphasis label and features of Japanese advertisement sentences. Experiments confirm that emphasized accent phrase prediction using support vector machine (SVM) offers encouraging accuracies for the system which requires emphasized accent phrase locations as context information to improve speech synthesis qualities.