Modelling the recognition of spectrally reduced speech

Jon Barker and Martin Cooke fj.barker,[email protected] Department of Computer Science, University of She eld, She eld, UK ABSTRACT Progress in robust automatic speech recognition may bene t from a fuller account of the mechanisms and representations used by listeners in processing distorted speech. This paper reports on a number of studies which consider how recognisers trained on clean speech can be adapted to cope with a particular form of spectral distortion, namely reduction of clean speech to sine-wave replicas. Using the Resource Management corpus, the rst set of recognition experiments con rm the high information content of sine-wave replicas by demonstrating that such tokens can be recognised at levels approaching those for natural speech if matched conditions apply during training. Further recognition tests show that sine-wave speech can be recognised using natural speech models if a spectral peak representation is employed in concert with occluded speech recognition techniques.