An audio processing apparatus includes a preprocessor which extracts a voice-band signal from a first electric signal, and outputs a first output signal containing the voice-band signal; a first controller which generates a first amplification coefficient for multiplying with the first output signal to compress a dynamic range of an intensity of the first output signal, and generates a first modified amplification coefficient by smoothing the first amplification coefficient with a first time constant; and a first multiplier which multiplies the first modified amplification coefficient and the first output signal. The first time constant is a first rise time constant when the intensity increases, and is a first decay time constant when the intensity decreases. The first rise time constant is not less than a temporal resolution of hearing of a hearing-impaired person, and is less than a duration time of sound which induces recruitment in the hearing-impaired person.