Speech recognition, artificial intelligence and translation: how rosy a future?
Henry S. Thompson · 1990
— The signal is lousy – it is noisy, and talkers are sloppy and lazy (Figure 1). The schwa (annotated as @) in the first syllable of potato is barely there at all – only about 20 milliseconds long, with much reduced excursion of the diagnostic properties. In general, talkers put no more information into the signals they produce than absolutely necessary. Hearers must use all sorts of ‘filters’ to reduce the resulting uncertainty of analysis.