Zoom out Search Issue

ManualsBrandsContents Manualsaudio & home theatreZoom in

IEEE SIGNAL PROCESSING MAGAZINE [19] MARCH 2015

listening devices (e.g., hearing aids or mobile phones) and allow to

exploit the spatial diversity in addition to the spectrotemporal

diversity, multichannel algorithms appear to be decisive for cur-

rent and future ALDs. Moreover, in contrast to single-microphone

signal enhancement algorithms, which have not been shown to

improve speech intelligibility but may reduce, e.g., the listening

effort, multimicrophone signal enhancement algorithms are capa-

ble of increasing speech intelligibility [1], especially when the

sound sources have different spatial characteristics.

Although microphone array signal processing, e.g., for telecon-

ferencing systems, is a well-established field dealing with similar

problems and signals [2], the problem setting for ALDs exhibits a

number of distinctive features. First, the microphone placement is

typically constrained by the fact that the devices should be incon-

spicuously placed at the user’s head and should capture the rele-

vant spatial information of the sound sources. Moreover, while

all signal enhancement algorithms ideally aim to remove the

undesired components and leave the desired components undis-

torted, the compromises need to be chosen differently depend-

ing on the application domain: for ALDs, distortion of the

desired signal or annoying noise artifacts will typically be penal-

ized more than a higher level of residual undistorted noise, and

the balance between reduced listener fatigue, increased speech

intelligibility, and subjective quality plays an even greater role

than in other speech communication devices. Finally, for bin-

aural systems that are expected to dominate the future markets,

preservation of the critical binaural cues as necessary for a cor-

rect spatial perception is crucial [3], not just for the desired sig-

nal, but also for the residual noise and interferers.

SCOPE

In this article, we will discuss several algorithms for multimicro-

phone signal enhancement and presentation that are suitable for

ALDs. The considered acoustic scenario is defined by a single

source of interest (target source) at any point in time, while mul-

tiple interfering point sources (e.g., competing speakers) and

additional incoherent noise (e.g., sensor noise, diffuse back-

ground noise) may be active simultaneously (see Figure 2). It is

assumed that some knowledge is available to distinguish the tar-

get source from the interfering sources once they are sufficiently

enhanced or separated. Bearing in mind that the wearers of ALDs

may move their heads, the relative positions of both the target

source as well as the interfering sources must be considered as

time-varying, so that source localization and tracking is required.

The fundamental concept of all considered multimicrophone

algorithms relies on spatial and/or spectrotemporal diversity,

i.e., the desired components should be separated from the

undesired components in the spatial and/or time-frequency

domain. The algorithms hence correspond to spatial filtering

[FIG1] The main processing blocks in an ALD.

[FIG2] A scenario with the target source (),st

point-like

interferers ( ),st

incoherent noise sources, and microphones at

the user’s head.

Signal Acquisition Signal Enhancement Signal Presentation

Monaural

or Binaural

Output

Right

Left

x(k, l )

w(k, l )

Out Right

Out Left

Source Position and

Binaural Cues

Localization/Binaural

Cue Determination

Target SourceTarget Source

InterferenceInterference

Incoherent

Noise

(t )

p,m

(t )

ALD User

THE WORLD’S NEWSSTAND

THE WORLD’S NEWSSTAND