Using Real Words for Recording Diphones

Abstract

This paper focuses on the creation of word-lists for making diphone recordings for speech synthesis. Such lists often consist of nonsense words, which has the advantage that the phonetic environment can be constrained, and it is easy to produce lists containing all possible combinations. However, this approach has the disadvantage that nonexperts may find it difficult to read the nonsense-word transcriptions. For this reason, we investigate here the issues associated with the use of real words in creating diphone recordings.