To generalize the assessment of speech recognizers a method has been developed to describe the performance of a recognizer for different degrees of variation of speech production and environmental parameters. For this purpose various speech parameters of minimal-difference words of a small database are manipulated by an analysis-resynthesis procedure. The recognizers are tested with this speech. To obtain a relevant range of manipulation, natural speech tokens have to be analysed. In this paper some results are presented of the analysis of natural speech and of the application of the obtained data in testing three commercially available recognizers. Also the influence of the structure of the test words is considered.
Cite as: Velden, J.G.v., Steeneken, H.J.M. (1989) RAMOS i: recognizer assessment by manipulation of speech. Proc. Speech Input/Output Assessment and Speech Databases, Vol.2, 139-142
@inproceedings{velden89_sioa,
author={Jeroen G. van Velden and Herman J. M. Steeneken},
title={{RAMOS i: recognizer assessment by manipulation of speech}},
year=1989,
booktitle={Proc. Speech Input/Output Assessment and Speech Databases},
pages={Vol.2, 139-142}
}