Using continuous lexical embeddings to improve symbolic-prosody prediction in a text-to-speech front-end | IEEE Conference Publication | IEEE Xplore