Speech recognition using peak frequency representations
Alan Crowe · 1990
Signal processing for speech recognition combines two distinct elements, spectral estimation, and feature selection. The author separates the issues of estimating the spectrum from that of describing it to a speech recogniser. He focuses on providing a description in terms of the frequencies of peaks. A minimum squared error criterion is used for specifying the description, but it is entirely distinct from that underlying LPC analysis, and is optimised by a dynamic programming technique. >