Schwa variants in american English

In this study, we examine the acoustic structure of schwa variants in a speech corpus intended for concatenative TTS. Our goals are two-fold. First, as a matter of academic interest, we seek to characterize schwa acoustics in a speech corpus designed to provide broad coverage of diand tri-phone contexts. This characterization extends recent work on the distinction between the barred-i variant of schwa /i/ and the more central /ә/ form of schwa [1]. Our second goal is to improve the concordance between our transcriptions as generated by the front-end of our TTS system, and the phonetic behavior of talkers who record corpora for concatenative TTS. Based on analysis of a single talker’s corpus, our results support the claim in [1] that /i/ is more common than generally assumed. However, the claim that /i/ characterizes all stemmedial schwas is not well-supported by our data.