Interspeech'2005 - Eurospeech
The most obvious difference between recordings of German words and non-sense words (logatomes) within the BITS project was an audibly noticeable difference in pitch: diphones obtained from logatomes seemed to have a higher pitch than those from words. We proved this by measuring the pitches of diphones concatenated to sentences, and by comparing the pitches of actually uttered logatomes with pitches of words. The results show, that the prompted material has influence on pitch values, i.e. that the design of corpora is a non-trivial problem; besides this, logatomes revealed higher pitch values.
We refer to Ellis' and Young's model of human language processing and speech production to support our hypothesis, that a higher mental workload during articulation of logatomes results in a rise of pitch. The question arises whether the usage of logatomes for obtaining high quality diphone synthesis is appropriate.
Bibliographic reference. Reubold, Ulrich / Steffen, Alexander (2005): "Pitch-effects in diphone recording: are logatomes inappropriate?", In INTERSPEECH-2005, 2797-2800.