Zoom out Search Issue

ManualsBrandsContents Manualsaudio & home theatreZoom in

IEEE SIGNAL PROCESSING MAGAZINE [78] MARCH 2015

running spectral analysis and the distribution of current levels

across electrodes is determined such that the loudness experi-

enced by the CI user is similar to that experienced by an average

listener with NH. Preliminary perceptual studies with CI recipi-

ents using SpeL confirmed that the relation between loudness

and the level and bandwidth of sounds was closer to normal [18].

More recently, SCORE (“Stimulus Control to Optimize Recipi-

ent Experience”), a simplified version of SpeL, was developed that

uses the estimated specific loudness function to calculate the

total loudness of sound signals in real time. Tests of speech recog-

nition with SCORE showed small but significant improvements

over ACE. Tests with an extended version for CI recipients who

use an acoustic hearing aid in the nonimplanted ear (SCORE

bimodal), suggested that it may improve speech recognition and

the ability of users to localize sounds, presumably because the

loudness differences between ears that carry information about

the direction of a sound source are conveyed more consistently

(cf.[19] and the references therein).

ENVELOPE ENHANCEMENT

In a CI, the electrical stimulation directly generates action poten-

tials in auditory neurons, predominantly bypassing any remain-

ing hair cells and synapse function. The synapse normally

demonstrates neural short-term adaptation [20], i.e., an increased

firing rate at the onsets of sounds. This short-term adaptation

acts as an across-channel phonological timing cue [20] and, with

conventional schemes such as CIS, is not present in the electri-

cally stimulated auditory nerve as in the normal auditory system.

Furthermore, recent studies have demonstrated that the tran-

sient parts of the speech envelope carry information that is

important for speech intelligibility in NH listeners.

Based on this rationale and former investigations, the

enhanced envelope (EE) strategy was developed and its feasibility

studied for applications in auditory prostheses. In this approach,

an additional processing stage is introduced after the envelope

detection stage (see Figure 2, block 11) wherein peaks, as a model

for the short-term adaptation and dependent on the onset rise

time, are added at the onsets in the envelope. This scheme is com-

plementary to the main structure of ACE or CIS.

The EE algorithm was evaluated with CI users and all lis-

teners demonstrated an immediate benefit with EE relative to

ACE [21]. The advantage of this enhanced envelope coding is

due to the emphasis of across-channel temporal coherence in

the coded speech signal. This temporal marker is an important

attribute for speech understanding in adverse listening situa-

tions and for sound source segregation; see also the electrodo-

grams in Figures 4–6. The onset enhancement is particularly

noticeable for the “b” sound in the word “boy” in Figure 5.

PERIODICITY MODULATION ENHANCEMENT

From psychophysical studies it is known that periodicity cues are

better perceived when modulation depth is high and modulations

are synchronized across channels to some extent [15]. This is

probably due to spread of excitation: electrodes close together

stimulate overlapping populations of neurons, which therefore

receive the aggregate stimulation pattern of multiple electrodes.

So if modulations are not synchronized across electrodes, the

modulation depth at the neural level may be severely reduced.

From the electrodograms in Figures 4–6, it is clear that

with most commercial strategies temporal modulations are not

well coded. In some channels, modulation depth is quite shal-

low and the desynchronization across channels combined with

spread of excitation leads to reduced modulation depth or even

spurious modulations in the aggregate pattern that will be

received by the auditory nerve fibers.

To improve this, several strategies have been developed

(e.g., [22]–[25] and the references therein). While the signal

processing to achieve it may differ, these strategies either

expand modulation depth or remove existing modulations and

explicitly modulate the envelope at the rate of F0. As an exam-

ple, in the following, the F0 modulation (F0mod) and eTone

strategies are briefly described.

The F0mod strategy is a simple example of a periodicity

enhancement strategy based on the ACE strategy. For each

frame of samples it estimates F0 and voicing probability using

an autocorrelation approach. If a frame is unvoiced, ACE pro-

cessing is applied. If a frame is voiced, all channels are modu-

lated synchronously using a sinusoidal modulator constructed

based on the F0 estimate. The block diagram and the output of

F0mod are shown in Figure 2 and Figures 4–6, respectively.

The eTone strategy [23] is based on the same principles but

includes an F0 estimator based on harmonic sieves, which is

very precise and robust to noise, and the modulated envelope is

mixed with the original envelope with a ratio depending on an

estimate of harmonicity of that particular channel. Modula-

tions are synchronous across channels and an exponential

decay modulation shape is used.

The F0mod and eTone strategies were evaluated for music

perception and speech recognition, and with tonal languages in

which pitch determines the lexical meaning of certain pho-

nemes (see [22]–[25] and the references therein). While period-

icity enhancement strategies can clearly improve periodicity

pitch perception, performance is still well below that of NH lis-

teners. For good pitch perception, listeners need access to all

three physical cues (see the “Introduction” section) and spectral

(place) and temporal cues need to be consistent. There are no

current CI strategies that make this possible, and we hypothe-

size that with the current electrode design and stimulation par-

adigm it is not possible to provide sufficiently place-specific

stimulation to achieve performance similar to NH. Note that for

a good representation of temporal information, good place spec-

ificity is required as well: when a population of neurons is stim-

ulated by information from several channels due to spread of

excitation, the aggregate pattern will be coded.

BILATERAL STRATEGIES

In various studies with controlled stimulation in laboratory

conditions, it has been found that bilateral CI users can be sen-

sitive to ITDs [26]. ITD thresholds, i.e., the smallest ITD that

can be detected, vary widely across subjects, with the best

THE WORLD’S NEWSSTAND

THE WORLD’S NEWSSTAND