TY - JOUR AU - Birch, G.E. AU - Zwierzynski, D.A. AU - Lefebvre, C. AU - Starks, D. PY - 1992/09/01 Y2 - 2024/03/28 TI - Continued development of an IMELDA based voice recognition system for persons with severe disabilities JF - Canadian Acoustics JA - Canadian Acoustics VL - 20 IS - 3 SE - Proceedings of the Acoustics Week in Canada DO - UR - https://jcaa.caa-aca.ca/index.php/jcaa/article/view/728 SP - 45-46 AB - The general design of the authors' speech recognition system is such that the acoustic features are first extracted from the speech signal in the front-end component. Subsequently, these features are passed on to the back-end component where they are compared with stored templates through the technique of dynamic programming. The front-end processing in their speech recognition system is a mel-scale fast-fourier-transform based spectral filter-bank analysis followed by a linear transformation. This linear transformation, which is called IMELDA, was developed and tested and is based on linear discriminant analysis. Comparisons with other systems indicate that it is the `state of the art' for front-end processing in a robust speech recognition system, outperforming other transforms and recognition systems, particularly in degraded speech ER -