Zoom out Search Issue

ManualsBrandsContents Manualsaudio & home theatreZoom in

IEEE SIGNAL PROCESSING MAGAZINE [75] MARCH 2015

Figures 4–6 show that ACE represents some speech formant

peaks and formant trajectories (i.e., changes in formant frequency

over time) more distinctly than CIS, particularly when background

noise is present. Because frequency bands containing relatively low

signal levels are not represented in the stimulation pattern, ACE can

enhance certain spectral features when perceived by CI users. This

may be one reason that several studies of speech understanding have

demonstrated slightly higher scores for ACE than CIS. For example,

Skinner et al. [10] reported that CI listeners in two separate compari-

son studies scored about six to nine percentage points higher, on

average, in sentence tests when using ACE rather than CIS.

FSP

Although most CI users obtain good performance with sound pro-

cessing schemes such as ACE and CIS, unfortunately intelligibility

of speech in competing noise is often unsatisfactory, and essential

components of musical sounds—particularly pitch—are poorly

perceived. Part of the reason may be the lack of TFS in the stimula-

tion patterns. In general, TFS is characterized by the rapid ampli-

tude variations within each of the band-pass filters that implement

the initial spectral analysis of sound signals. In contrast, only the

slowly varying envelope of the band-limited signals is used to mod-

ulate stimulation levels in schemes such as ACE and CIS.

In the quest for improved CI sound processing, numerous

attempts have been made to introduce TFS cues explicitly. One

such scheme, currently the default in systems manufactured by

Med-El, is known as FineHearing Technology. The aim of Fine-

Hearing Technology is to represent TFS information present in the

lowest frequencies of the input sound signals by delivering bursts

of stimulus pulses on one or several of the corresponding CI elec-

trodes. These bursts can consist of one or more stimulation pulses

and are derived indirectly from the band-limited acoustic signals.

Each burst is triggered by a positive zero-crossing in the bandpass-

filtered waveform, while stimulus pulses within the burst are deliv-

ered at a constant, high rate that depends on user-specific settings

(typically 5,000–10,000 pps). The duration and amplitude-envelope

modulation of each burst are predetermined to approximate the fil-

tered acoustic waveforms after half-wave rectification. These bursts

contain information about the TFS in the lower frequency bands

that is not available in the envelope of those signals, potentially

leading to improved perception for CI users. In essence, FineHear-

ing Technology uses variable-rate coding to provide additional

information about the TFS of the signal. Med-El has released the

FSP, FS4, and FS4-p coding strategies. These strategies differ

mainly in the frequency range across which TFS is presented.

While in FSP, TFS is represented for frequencies up to 350–500 Hz;

in FS4 and FS4-p, TFS is presented for frequencies up to 750–950 Hz.

To faithfully represent F0, these strategies cover an input frequency

range from 100–8,500 Hz by default, which differs from the CIS

strategies from Med-El (250–8,500 Hz). The FSP coding strategy is

illustrated in Figures 3–6, where TFS pulse patterns are delivered

by the two most apical electrodes while the remaining electrodes

convey CIS-like pulse trains.

Several of the coding strategies available in the Med-El system

have been compared in a number of studies. Most published

studies evaluating the perception of CI recipients when using FSP

relative to other sound processing schemes (e.g.,CIS) are difficult

to interpret. In some cases, the sound-processor hardware and

settings such as the input frequency range were altered at the

same time as the processing algorithm was changed. In one study

of 46 experienced CI users where such differences were explicitly

taken into account, no significant differences were found between

FSP and a variant of CIS in speech perception tests, although the

participants’ subjective preferences generally favored FSP [11].

Moreover, it should be noted that in some experiments the fitting

of the CI system to recipients was not altered when changing

from CIS to FSP. The study by Riss et al.[12] seems to indicate

that at least some of the short-term improvements that have been

reported with FSP can be attributed to the extended frequency

range. As studies with the newer FS4 and FS4-p strategies are

ongoing, further research is needed to quantify perceptual out-

comes more thoroughly.

HiRes120

Another sound processing scheme designed to enhance delivery

of TFS information to CI recipients is used in systems manufac-

tured by Advanced Bionics. Known as HiRes120, this scheme

applies a technique to identify the dominant spectral peak within

each of the band-pass filters that perform the spectral analysis of

incoming sounds. The frequency of each spectral peak is used to

control a synthetic modulator such that the modulations contain

temporal information derived from each frequency band that is

not present in the amplitude envelope of the band-limited signals

[13]. These modulations are combined with the corresponding

envelope levels and then sampled in synchrony with the pulses

delivered to the electrodes. The typical pulse rate on each elec-

trode is about 2,000 pps. At the same time, the estimated peak

frequency within each of the analysis filters is used to control the

relative currents of pulses delivered simultaneously on two adja-

cent electrodes that are allocated to the filter. There are 16 intra-

cochlear electrodes in the Advanced Bionics implant, and

therefore 15 paired electrodes can be allocated to the filters. By

varying the relative currents on the electrode pairs, so-called vir-

tual channels are created, and it is assumed that the site of maxi-

mal neural activity can be steered with finer spatial resolution

than is possible when the electrodes are activated one at a time.

With HiRes120, eight different ratios of current are implemented,

leading to eight virtual channels per adjacent pair of physical

electrodes. HiRes120 is claimed to provide improvements over

sound processing schemes such as CIS in both temporal and spa-

tial resolution of the stimulation patterns. The main differences

between these stimulation schemes are most clearly visible in the

electrodograms of Figures 3 and 5. Additionally, a graphical rep-

resentation of the virtual channels is shown in Figure 7.

HiRes, which is a CIS-like strategy without current-steering,

has been compared with HiRes120 in various studies (e.g., [14]

and the references therein) using measures of speech perception

in quiet and in noise, and music perception. There were no clear

significant effects of the processing strategy on any of the speech

and music perception abilities nor on temporal modulation

THE WORLD’S NEWSSTAND

THE WORLD’S NEWSSTAND