Selecting non-uniform units from a very large corpus for concatenative speech synthesizer | IEEE Conference Publication | IEEE Xplore