Automatic generation of synthesis units for trainable text-to-speech systems | IEEE Conference Publication | IEEE Xplore