Voice Modification for Applications in Speech Synthesis

A significant part of the work required to create a high quality speech synthesizer is the creation of “synthetic voices”. Reusing an existing voice database and making it sound like a different speaker, or like the same speaker in a different emotional state, or using a different speaking style, is obviously important for increasing the efficiency in creating voice options for a synthesizer. This article reviews techniques to change signal characteristics like pitch and durations, and also spectral modifications. We conclude by assessing the prospects of voice modification in speech synthesis in light of nowavailable advanced machine learning techniques.