A method is proposed to specify the performance of a recognizer as a function of the variation of specific speech parameters and environmental conditions. The method uses a small test vocabulary with minimal difference wordsets of CVC-type words. Three wordsets are available including all Dutch initial consonants, vowels, and final consonants, respectively. This allows all possible confusions. By means of an analysis-resynthesis method, the testwords will be physically manipulated according to the changes of human speech in defined conditions. In an earlier paper (6), the results of a pilot study based on recognition experiments with this method were presented. In this paper the evaluation of the test vocabulary and the analysis of representative speech tokens are considered. The speech tokens include: male and female speech, inter- and intra-speaker variation and speech recorded just after a physical stress.
Bibliographic reference. Steeneken, Herman J. M. / Velden, Jeroen. G. van (1989): "RAMOS - recognizer assessment by means of manipulation of speech", In EUROSPEECH-1989, 2316-2319.