The human ear remains the best speech recognizer and should anchor machine systems

Evidence

The human hearing system is still, in every respect, by far the best speech recognition system. Therefore, it is reasonable to simulate this system as much as possible. Such a simulation should be based on available physiological and psychoacoustical knowledge. In addition, linguistic and phonetic rules should be introduced into the recognition procedure. … It contains the nonlinear peripheral preprocessing with active feedback, followed by the extraction of basic auditory sensations out of which complex auditory sensations such as virtual pitch or rhythm may be created. All of these sensations are checked for dominant variations. The speech recognition procedure also makes use of non auditory information such as linguistic and phonetic rules and finally produces a sequence of phonetic items. — Professor Dr.-Ing. Hugo Fastl, Professor Dr.-Ing. Eberhard, Psychoacoustics- Facts and Models, 2027

Categorization

• DEFINE: Outlines a full psychoacoustic speech-recognition architecture: nonlinear peripheral preprocessing, auditory sensation extraction, then linguistic rules.
• XREF: Ties to engineering/neuroscience bridges — cochlear-model speech recognition systems.