![]() |
ESCA Tutorial and Research Workshop on
|
![]() |
Within the research field of synthesis assessment methodologies in which the ESPRIT-SAM project is involved, three tests were simultaneously defined and ran in three European laboratories, evaluating English, French and German languages [see related communications of Hazan & Grice and of Jekosh in this workshop]. Similar corpora were used in these languages, following SAM decisions on "Semantically Unpredictable Sentences" (SUS) [van ERP and Grice, 1989]. The French test here presented involved twenty SUS per syntactic structure. Hundred sentences were generated under five conditions : two coding techniques, both under two prosodic models, and natural speech. The four synthesizers used the same diphones dictionary, obtained from the voice of one speaker which represented the reference natural speech.
We were therefore able to compare the "same voice", synthesized with the same diphones concatenating method under varying aspects :
Besides, to evaluate the linguistic influence of the corpus on this test methodology, we also compared listeners' answers on "handily semantized" sentences vs. randomly generated ones and on "feed-forwarded" (e.g. preknown) sentences vs. unknown ones.
Bibliographic reference. Benoît, Christian (1989): "Intelligibility test for the assessment of French synthesisers using semantically unpredictable sentences", In SIOA-1989, Vol.2, 27-30.