Speech recognition

Discusses how successful speech recognition comes about through the careful engineering of a complete system. This requires the appropriate separation and application of speech knowledge, pattern processing techniques, optimisation techniques, and spectral estimation. There is still much work required on all aspects of the speech recognition problem. However, from the spectral estimation point of view the major challenges lie in the identification and extraction of features that are robust to noise and that are invariant to coarticulation, intraspeaker variation and interspeaker variation. These are not simply problems of better spectral estimation but also problems of what features to extract from the incoming speech