Prosody Prediction from Syntactic, Lexical, and Word Embedding Features

ROSE SLOAN; RITVIK SHRIVASTAVA; SYED SARFARAZ AKHTAR; AGUSTÍN GRAVANO; BRYAN LI; JULIA HIRSCHBERG

Viena

Workshop; 10th ISCA Speech Synthesis Workshop; 2019

International Speech Communication Association

Accurate prosody prediction from text leads to more natural-sounding TTS. In this work, we employ a new set of features to predict ToBI pitch accent and phrase boundaries from text. We investigate a wide variety of text-based features, including many new syntactic features, several types of word embeddings, co-reference features, LIWC features, and specificity information. We focus our work on the Boston Radio News Corpus, a ToBI-labeled corpus of relatively clean news broadcasts, but also test our classifiers on Audix, a smaller corpus of read news, and on the Columbia Games Corpus, a corpus of conversational speech, in order to test the applicability of our model in cross-corpus settings. Our results show strong performance on both tasks, as well as some promising results for cross-corpus applications of our models.