An infrastructure for Turkish prosody generation in text-to-speech synthesis

Text-to-speech engines benefit from natural language processing while generating the appropriate prosody. In this study, we investigate the natural language processing infrastructure for Turkish prosody generation in three steps as pronunciation disambiguation, phonological phrase detection and intonation level assignment. We focus on phrase boundary detection and intonation assignment. We propose a phonological phrase detection scheme based on syntactic analysis for Turkish and assign one of three intonation levels to words in detected phrases. Empirical observations on 100 sentences show that the proposed scheme works with approximately 85% accuracy.