A problem of state-of-the-art TTS systems is that the produced speech did not always fits the actual speech task. An improvement can be achieved by considering speaking styles in synthetic speech. If the desired style deviates only slightly from the standard speech of the TTS system, as it is the case for different kinds of reading styles, it is proposed that the style can be simulated with adapted prosody. Therefore this investigation uses data driven algorithms for prosody generation. After training style simulation is done by switching to the appropriate style parameter set. First experiments show that the resulting speech quality is increased by the style adapted prosody.
展开▼
