Evaluation

The Arabic voice was evaluated with Diagnostic Rhyme (DRT)[8] and Modified Rhyme (MRT) tests, and with a sentence-level test in which listeners were asked to mark any words that sounded bad. The DRT asks listeners to choose which of a pair of monosyllabic words that differ only in one feature in the word-initial phoneme is being synthesized. For example, a test for voicing might include the words ``tart'' and ``dart.'' The MRT presents the listener with a list of monosyllabic words that differ in either the first or last phoneme, asking them to choose which word is being synthesized.

Standard DRT distinctive feature categories are defined for English and include voicing, nasality, sustenation, sibilation, graveness and compactness. The DRT and MRT categories were adapted to include important distinctions in Arabic; emphaticness was added as a diagnostic class. Gemination was not tested; it was our experience that speakers varied considerably in their perception of geminate consonants.

The score of a test was defined to be

$$\frac{\mbox{number of items guessed correctly}}{\mbox{total number of items}}$$

Results are shown in Table 1, and are comparable to English voices of the same size and degree of development.

Table 1: Evaluation results for the Arabic voice

DRT	MRT	Sentence
78.3	72.0	84.7