Should recognizers have ears

摘要: Abstract Recently, techniques motivated by human auditory perception are being applied in main-stream speech technology and there seems to be renewed interest implementing more knowledge of communication into a design recognizer. The paper discusses the author's experience with applying automatic recognition speech. It advances notion that reason for such engineering should ability suppress some parts irrelevant information message argues against blind implementation scattered accidental which may task. following three properties discussed detail: • limited spectral resolution, use from about syllable-length segments, ignore corrupted or components It shows referring published works selective knowledge, optimized on cases derived real data, can consistent current stochastic approaches ASR could yield advantages practical applications.