speech processing IN cochlear implants

The cochlear implant speech processors have to adapt the stimulation of the inner ear to the properties of the electrical auditory nerve responses. These include:

  1. 1.Reduction of the spectral content to the number of electrodes that can provide independent stimulation to the nerve. This means that the number of stimulated sites is reduced. Typical value is ~12.

  2. 2.Channels have to be mapped to the cochlear frequency representation; this is logarithmic (4.4 mm/octave in humans).

  3. 3.Stimulation is with biphasic, charge-balanced pulses. The dynamic range of electrical stimulation is less than that with acoustic stimulation of the hair cell/auditory nerve complex. Consequently, compression functions recalculate the amplitudes.

  4. 4.Stimulation is interleaved between channels to avoid current interactions in the cochlea (interleaved sampling of the speech).

  5. 5.In some devices, to further minimize channel interactions, the stimulation order along the cochlea is pseudo-randomized so that neighbouring electrodes are not activated in close succession.

  6. 6.Some companies use several current sources in their devices, allowing current steering procedures and higher stimulation rates.


Speech is characterized by

  1. 1.Temporal variations of sound pressure (amplitude modu-lations)

  2. 2.Temporal variations of frequency composition (modu-lations)

  3. 3.Characteristic spectral peaks (vowels)

Top: the spectrum of a vowel with fundamental frequency (*) and the first 2 formants. Below, the frequency of these formats are shown for different vowels („CV system“). Some vowels overlap in their frequency components and the interindividual variability is very high.

cochlear implants

Block diagram of the continuous interleaved sampling method (Wilson et al.,  1991, Nature) used by all cochlear implants in modified versions (so-called N-of-M strategies). The signal is split into the frequency channels used to stimulate the nerve by bandpass filters. Second, rectification and low-pass filtering is performed to extract the envelope of the signal in each channel. Compression adapts the dynamic range. Note that individual channels are stimulated in succession (not simultaneously)