The ESCA Workshop on Speech Synthesis

September 25-28, 1990
Autrans, France

A New Algorithm for a Concatenative Speech Synthesis System Using an Augmented Acoustic Inventory of Speech Sounds

Joseph P. Olive

AT&T Bell Laboratories, Murray Hill, NJ, USA

Previously we discussed a speech synthesis by rule scheme that consisted of concatenating small elements of analyzed natural speech segments. These segments consisted of transitions between adjacent phonemes and were stored in terms of LPC derived area parameters. Although the speech produced from this scheme was highly intelligible, it did not sound natural or continuous. Investigation showed that most short and reduced vowels, were not described correctly by the previous method, because they depended too much on their environment. Depending on their neighbors, often, these phonemes do not reach their target and thus can not be defined by diphonic units. Recently, we have introduced a scheme that can access a larger variety of acoustic inventory elements consisting of the previously described transitions as well as longer units. The longer multiphonic units consist of triphones for the short vowels and many common words, especially small function words. Due to the new inventory, the speech produced by the new synthesis scheme has maintained its high intelligibility, but sounds more continuous and pleasant.


Full Paper

Bibliographic reference.  Olive, Joseph P. (1990): "A new algorithm for a concatenative speech synthesis system using an augmented acoustic inventory of speech sounds", In SSW1-1990, 25-30.