The test material was recorded by 48 test speakers without making an a priori division of speakers into the development test and evaluation test groups. Two evaluation sets and a development set of speakers were defined, with the aim of obtaining balanced speaker groups. The procedure that placed speakers in the different test sets aimed to balance the number of male and female speakers in each set, then to balance the age distribution, and finally a coarsely calculated speaking rate measure. Therefore, each of the 3 test groups contain roughly the same proportion of speakers of the same sex and from the same age range. The speaking rate was computed for each test speaker by dividing the total number of samples per speaker by the number of words in the transcription .dot files. The adaptation material was not included in this computation as it was missing for one speaker. Within each age/sex group individual speakers with comparable speaking rates were then equally distributed over the test groups.
The development test speaker group comprises 18 speakers and is distributed with the training material. The two evaluation test sets contain data from 14 speakers that will be released at a later stage.