The invention provides a voice
processing method applied in an electronic ear. The method comprises the steps of carrying out pre-emphasis treatment on an input voice
signal by promoting
high frequency components, then dividing the voice
signal into m frequency bands through a filter group consisting of m band-pass filters, obtaining envelope signals of m channels by full-wave rectification and low-pass
filtration, calculating the modulation depth of each channel, selecting n channels with greatest modulation depth, carrying out non-linear function compression on envelopes of the n channels for obtaining a relatively narrow
dynamic range, using a symmetric dual-phase
pulse sequence for carrying out modulation on envelope amplitude information after the compression of the n channels, modulating the asynchronization of the
pulse sequence on
time sequence, appearing alternating pulses and eliminating interferences among the channels. The
stimulation rate of each channel obtained by usingthe n channels for stimulating an
electrode is larger than the
simulation rate of each channel when using m channels for simulating the
electrode under the condition of constant total
stimulation rate, thereby transferring more
time domain details of the voice
signal and improving the accuracy of voice recognition.