22.03.2013 Views

Study report SoundSmoothing - Siemens Hearing Instruments

Study report SoundSmoothing - Siemens Hearing Instruments

Study report SoundSmoothing - Siemens Hearing Instruments

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

www.siemens.com/hearing<br />

<strong>Study</strong> <strong>report</strong><br />

<strong>SoundSmoothing</strong><br />

Despite such modern features as multichannel, wide dynamic range compression (WDRC),<br />

output limiting and noise suppression, many hearing instrument users still <strong>report</strong> discomfort<br />

from common transient noises like paper rustling or clattering dishes (Hernandez et al., 2006).<br />

In keeping with the company‘s innovative, user-oriented philosophy, <strong>Siemens</strong> has developed<br />

<strong>SoundSmoothing</strong>®, an algorithm that substantially reduces impulsive, non-speech sounds. Recently,<br />

the National Acoustic Laboratories (Australia) conducted an interesting and significant<br />

clinical study to evaluate the effects of this new algorithm (O'Brien et al., 2006).<br />

<strong>SoundSmoothing</strong>: How it works<br />

After a spectro-temporal analysis of the microphone signal, envelope features are extracted. These<br />

are then used to determine whether speech or non-speech sounds are present. To accomplish<br />

this, the envelope features are analyzed using a speech model. Then an attenuation factor is<br />

calculated for non-speech sounds only. The amount of transient gain modification is determined<br />

by the ratio of peak level to long-term overall RMS level (i.e. the more transient, the more gain<br />

reduction).<br />

Fig. 1 Algorithmic implementation of <strong>SoundSmoothing</strong><br />

Objectives<br />

The objectives of this study were to determine if:<br />

• <strong>SoundSmoothing</strong> provides a noticeable benefit to hearing aid users when listening to<br />

transient noises.<br />

• <strong>SoundSmoothing</strong> negatively affects speech perception and / or the horizontal localization<br />

of sounds containing transients.<br />

Evaluation of a transient noise reduction algorithm<br />

Josef Chalupper and Beate Krämer<br />

Corporate Audiology, <strong>Siemens</strong> Audiologische Technik GmbH, Germany<br />

Speech model<br />

Input Spectro-temporal Extraction<br />

Detection Transient gain Resynthesis Output<br />

analysis<br />

of envelope of non-speech modification<br />

features<br />

components


Methods<br />

21 hearing-impaired subjects ranging in age from 24<br />

to 85 years, with a median age of 75, participated in<br />

the study. All subjects had a symmetrical sensorineural<br />

hearing loss. The mean three-frequency average<br />

(3FA) hearing loss of the group ranged from 26 to<br />

58 dB HL, with a mean 3FA of 43 dB HL. The subjects<br />

had worn amplification devices from 1.5 to 21 years,<br />

with a mean of 7.5 years. All subjects were fitted<br />

binaurally with CENTRA P BTE housings containing<br />

microphones and receivers. The signal processing<br />

was done in an external computer. The devices were<br />

fitted according to NAL-RP (Byrne et al.1991). Linear<br />

amplification was chosen to separate the effect of<br />

<strong>SoundSmoothing</strong> from other signal processing strategies<br />

that adaptively change gain.<br />

The <strong>SoundSmoothing</strong> algorithm could be disabled or<br />

enabled on three strengths: minimum, medium, and<br />

maximum. The algorithm was evaluated using six<br />

stimuli comprising:<br />

• Speech within two loud, repeated transient<br />

sounds (door slamming and hammering nails)<br />

• Speech within two softer sounds with frequent<br />

amplitude fluctuations (paper rustling and cutlery<br />

clattering)<br />

• Speech within a medium level of stationary<br />

sounds (party noise)<br />

• Speech in quiet surroundings<br />

In a round-robin paired-comparison test, the three<br />

strengths and the “off” condition were compared<br />

five times for speech in the door slamming, paper<br />

rustling and quiet stimuli. For the remaining stimuli<br />

only the medium strength and the ”off” condition<br />

were compared. Each comparison was rated in terms<br />

of preference strength (slightly better, moderately<br />

better, or much better). For each stimulus, the<br />

subjects were asked to indicate what subjective<br />

listening criteria they had used to determine their<br />

preference.<br />

Horizontal localization performance was measured<br />

using the hammering nails noise as stimulus. Horizontal<br />

testing was conducted in an anechoic chamber<br />

with an array of 20 loudspeakers. A 1.5-second<br />

sample of the hammering nails noise was used as<br />

the test stimulus. The subjects were asked to verbally<br />

<strong>report</strong> the perceived direction of the stimuli.<br />

”A strong preference for<br />

<strong>SoundSmoothing</strong> can be seen<br />

for all impulsive noises.“<br />

Speech recognition was measured in the more continuous<br />

paper rustling and cutlery clattering noises.<br />

All three strengths as well as the ”off” condition<br />

were included in these tests. For the speech discrimination<br />

task, the Bamford-Kowal-Bench/Australian<br />

version (BKB/A) Standard Sentence Lists (Bench et al.<br />

1979) were administered.<br />

Prior to testing, the subjects were asked to what extent<br />

they were bothered by sudden sounds in their<br />

everyday lives.<br />

Results<br />

Preference for <strong>SoundSmoothing</strong><br />

For simplification, the results of the various Sound-<br />

Smoothing strengths were combined for each stimuli<br />

– i.e. “on” setting comprises “min”, “med” and “max”.<br />

A strong preference for <strong>SoundSmoothing</strong> can be<br />

seen for all impulsive noises, especially for the loud<br />

and repeated ones (hammering, door slamming).<br />

A two-sided binomial test revealed statistically significant<br />

preference for <strong>SoundSmoothing</strong> “on“ for<br />

hammering (p < 0.001), door slamming (p = 0.007)<br />

and paper rustling (p = 0.002). No preference can<br />

be seen for the stationary party noise and the speech<br />

signals. This is probably due to the fact that<br />

<strong>SoundSmoothing</strong> does not affect these signals and<br />

therefore the subjects did not hear any difference<br />

between the “on” and “off” condition, yet the test<br />

conditions forced the participants to choose between<br />

them. Nevertheless, this is a very positive finding, as<br />

it confirms that <strong>SoundSmoothing</strong> does not cause any<br />

artifacts or interferences that would reduce speech<br />

quality. Generally, the subjects used clarity of speech<br />

as their main listening criterion in the paired comparison<br />

test, although loudness and comfort of noise<br />

rated higher (or equally high) when listening to<br />

speech in the three most transient noises (door slamming,<br />

hammering nails, and cutlery clattering).<br />

Repeated measures ANOVAs on the weighted pairedcomparison<br />

preference scores were calculated for<br />

each stimulus, using grouping of the subjects by how<br />

bothersome they found sudden sounds as the between-subject<br />

variable. These analyses revealed no<br />

significant effect of bothersomeness (p > 0.05). This<br />

indicates that the degree to which sudden sounds<br />

bother subjects in their everyday life is not a good<br />

predictor of whether or not they will prefer Sound-<br />

Smoothing for the stimuli used in the present study.<br />

Horizontal localization<br />

Horizontal localization was tested with the conditions<br />

<strong>SoundSmoothing</strong> “on”, “off” and also with 10<br />

hearing-impaired subjects in an unaided condition<br />

(Keidser et al., 2006). The subjects <strong>report</strong>ed no significant<br />

effect of <strong>SoundSmoothing</strong> on horizontal<br />

localization performance in the left/right dimension<br />

(p = 0.33) or in the front/back dimension (p = 0.44).


Fig. 2 Preference for <strong>SoundSmoothing</strong><br />

Preference / %<br />

100<br />

80<br />

60<br />

40<br />

20<br />

0<br />

Hammering Door<br />

slamming<br />

Fig. 3 Speech intelligibility in<br />

noise with <strong>SoundSmoothing</strong><br />

SRT / dB SNR<br />

0<br />

-1<br />

-2<br />

-3<br />

-4<br />

-5<br />

-6<br />

-7<br />

-8<br />

-9<br />

-10<br />

Paper<br />

<strong>SoundSmoothing</strong> off<br />

<strong>SoundSmoothing</strong> min<br />

<strong>SoundSmoothing</strong> med<br />

<strong>SoundSmoothing</strong> max<br />

Cutlery<br />

Paper<br />

rustling<br />

Mean and 95% confidence intervals of speech reception<br />

threshold (SRT); research conducted at the National Acoustic<br />

Laboratories, Australia<br />

<strong>SoundSmoothing</strong> "on" <strong>SoundSmoothing</strong> "off"<br />

Cutlery<br />

clattering<br />

Speech Party<br />

noise<br />

Speech recognition<br />

Speech discrimination testing was conducted in the<br />

same anechoic chamber used for horizontal localization<br />

testing, as described above.<br />

The mean and 95% confidence intervals are shown<br />

in Figure 3. The subjects needed a more favorable<br />

(although still negative) signal-to-noise ratio to understand<br />

50% of the speech signal in paper rustling<br />

noise than in cutlery clattering noise. Analyses of variance<br />

were performed on the signal-to-noise ratios<br />

obtained for each noise using the <strong>SoundSmoothing</strong><br />

setting as a variable. There was no significant effect<br />

of <strong>SoundSmoothing</strong> on speech recognition in paper<br />

rustling (p = 0.65) or cutlery clattering (p = 0.97)<br />

noises.<br />

Percent of participants preferring<br />

<strong>SoundSmoothing</strong><br />

”on” versus ”off” for<br />

different sound samples<br />

using a paired-comparison<br />

procedure; research<br />

conducted at the National<br />

Acoustic Laboratories,<br />

Australia


Conclusions<br />

Although <strong>SoundSmoothing</strong> has its greatest effect on transient sounds with a high input level like door<br />

slamming or hammering, the subjects preferred <strong>SoundSmoothing</strong> “on” for all transient noises, including<br />

softer sounds with a frequent amplitude fluctuation (e.g. paper rustling, cutlery). There were absolutely no<br />

indications that <strong>SoundSmoothing</strong> has any negative effects on speech recognition or localization. The new<br />

CENTRA products are the first hearing instruments to be equipped with <strong>SoundSmoothing</strong>. It is expected that<br />

<strong>SoundSmoothing</strong> will soon be a standard feature in many high-end hearing instruments.<br />

References<br />

O‘Brien, A., Keidser, G., Convery, E., Dillon, H. (2006).<br />

Evaluation of a transient noise reduction algorithm.<br />

NAL internal <strong>report</strong>: NAL011.<br />

Hernandez, A. R., Chalupper, J., Powers, T. A. (2006).<br />

An assessment of everyday noises and their annoyance.<br />

The <strong>Hearing</strong> Review, 13(7), 16–20.<br />

Bench, R. J., Doyle, J., Daly, N., Lind, C. (1979).<br />

The BKB/A Speechreading (Lipreading) Test.<br />

Victoria: La Trobe University.<br />

Byrne, D., Parkinson, A., Newall, P., in:<br />

Studebaker, G. A., Bess, F. H., and Beck, L. editors. (1991).<br />

Modified hearing aid selection procedures for severe/profound<br />

hearing losses: Parkton, MD.<br />

York Press. p. 295–300.<br />

Keidser, G., Rohrseitz, K., Dillon, H., Hamacher, V.,<br />

Carter, L., Rass, U., Convery, E. (2006).<br />

The effect of multi-channel wide dynamic range compression,<br />

noise reduction, and directional microphone on horizontal<br />

localisation performance in hearing aid wearers.<br />

Int J Audiol., in press.<br />

<strong>Siemens</strong> Audiologische Technik GmbH (2006).<br />

<strong>SoundSmoothing</strong>: A new algorithm to reduce the annoyance<br />

of noise.<br />

<strong>Siemens</strong><br />

Audiologische Technik GmbH<br />

Gebbertstrasse 125<br />

91058 Erlangen<br />

Germany<br />

Phone +49 (9131) 308-0<br />

www.siemens.com/hearing

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!