Phonak Insight

April 2016

SoundRecover2 – the adaptive frequency

compression algorithm
More audibility of high frequency sounds
Phonak led the way in modern frequency lowering technology with the introduction of SoundRecover in 2008. Since then, extensive
worldwide field studies with adults and children have found increased detection, distinction and recognition of high frequency sounds,
better speech understanding and significant improvement in intonation and overall voice quality for users. For those with more extreme
severe-to-profound losses, however, including left corner audiograms and ski-slope losses, the benefits have been limited due to the
restricted audible bandwidth in which frequency compression could be applied. The new SoundRecover2 algorithm aims to restore the
audibility of relevant high frequency sounds while leaving intact the low frequency structures important for good sound quality. The new
frequency lowering scheme retains the essence of the original SoundRecover, utilizes an adaptive algorithm and an additional cut-off
frequency in order to successfully extend these benefits to those with more extreme severe-to-profound hearing loss.

Introduction
The remapping of frequencies for extending the perceptual auditory limited to a minimum value of 1.5 kHz. This ensures that important
bandwidth of hearing aid users has been commercially available for vowel structures are left unchanged, and at the same time allows
approximately 10 years now. Phonak introduced SoundRecover, presentation of the noise-like high frequency components of
non-linear frequency compression, with the first Naída in 2008, speech in the individual audible range. A more detailed discussion
offering a solution for restoring audibility of conventionally of the SoundRecover frequency lowering system is provided in the
unaidable high frequency sounds. paper by McDermott 2010.1

In order to achieve distortion-free amplification of the input signal, Experience has shown that the SoundRecover frequency lowering
SoundRecover exploits the fact that vowels are dominated by scheme works very well for speech and for high frequency sounds
greater energy in the low frequencies whereas voiceless fricatives such as bird song or environmental sounds. It can be successfully
are dominated by greater energy in the high frequencies. For this applied to hearing losses where aidable hearing is available above
reason, SoundRecover was designed with a cut-off frequency which 1.5 kHz allowing the high frequencies to be compressed into an
is the starting point of compression. For inputs below the cut-off audible region. However, the fitting of more extreme severe-to-
frequency, the input signal is not subject to frequency compression. profound losses, like left corner audiograms and ski slope
All inputs above the cut-off frequency are subject to frequency audiograms, where users have only low frequency hearing
compression. Hence, the output area below the cut-off frequency is thresholds, has presented unique challenges. These losses require
unchanged, while the output area between the cut-off frequency more aggressive parameter settings (lower cut-off, stronger
and the upper edge frequency is compressed with a constant compression ratio) than those admissible with SoundRecover due to
compression ratio1. The upper-edge frequency corresponds to the concerns about impact on sound quality.
maximum output frequency and is set according to the audible
bandwidth of the individual audiogram. The cut-off frequency is In order to widen the reach of SoundRecover, the new
SoundRecover2 algorithm is designed to allow operation with lower
1 cut-off frequencies and weaker compression ratios, thereby
The compression ratio is constant if viewed on logarithmically scaled
frequency axes. extending the benefits of frequency compression to a broader
audience of children and adults.
Functional Description of SoundRecover2
Basic functional principle
SoundRecover2 is an adaptive frequency lowering scheme based on
the original SoundRecover framework. The new algorithm protects
vowels which are dominated by lower frequency energy and
compresses voiceless fricatives which are mainly composed of
energies in the higher frequency range. Figure 1 depicts the
spectral distribution of different English phonemes.

Figure 2: Sample output curve for SoundRecover2. Depending on the energy

distribution of the input signal, the frequency compression starts either at the
lower or at the upper cut-off frequency.

Impact of the adaptive behavior

The impact of the adaptive behavior of SoundRecover2 is
substantial. The adaptively determined starting point of
compression ensures that input signal components are lowered, to
a high extent, only when they have significant high frequency
Figure 1: The speech banana showing the frequency content of speech sounds
in English. energy. As a result, the lower cut-off can be set well below the
current limit of 1500 Hz increasing the area of compression and
The significant difference with SoundRecover2 is that the extent of allowing in turn, weaker compression ratios than possible with the
frequency lowering, i.e. the area of protection and the starting original SoundRecover. The value of the upper cut-off remains
point of compression is not fixed, but is instead set adaptively as a limited however, and can be set quite high, because the upper cut-
function of the input signal. This adaptive nature is realized by the off is applied only when significant low frequency energy is present.
use of two cut-off frequencies, of which only one is active at any This extends the output area below the upper cut-off frequency
moment in time. Based on the momentary energy distribution in where the signal is protected and left intact. Thus it is guaranteed
the input signal, the system determines instantaneously which one that tonal structures and other low frequency components are
of the two cut-off frequencies is applied. Thus, the functional protected and not compressed, and in particular, it is ensured that
principle of SoundRecover2 is similar to that in SoundRecover; with important vowel formants will not be adversely affected.
the increased sophistication that it now switches automatically
between two possible starting points of compression, respectively Outcomes
between a “lower” and an “upper” cut-off frequency. As in The adaptive nature of SoundRecover2 allows frequency lowering
SoundRecover, frequency lowering is always carried out with a with a lower overall cut-off frequency and weaker compression
predefined constant compression ratio regardless of which cut-off ratio than was available with the original SoundRecover. The high
frequency is momentarily active. value of the upper cut-off yields a better sound quality with more
naturalness and familiarity, and less distortions of tonal
Hence, SoundRecover2 instantaneously maps input components to components. At the same time the lower possible value of the
the output depending on their energy content. This adaptive lower cut-off provides for extended access to high frequency
frequency lowering processing is accomplished by simply sounds for all kinds of hearing losses. Thus, the fitting range of
recognizing the different energy distributions of tonal and noise- SoundRecover2 is expanded to include subjects with a very
like structures of the input signal. In case of more low frequency restricted audible bandwidth as commonly seen with those who
content, frequency compression takes place with the upper (higher) have greater severe-to-profound hearing loss, left corner
cut-off frequency in order to “protect” the low frequency sounds audiograms and ski slope losses. Put simply, more users will be able
from being compressed. In case of more high frequency content, to benefit from frequency lowering than ever before with the new
frequency compression takes place with the lower cut-off SoundRecover2 algorithm. The use of a weaker compression ratio
frequency to restore audibility of the high frequency sounds. When results in less alteration and thus an improved preservation of the
applied to speech signals, this strategy leaves vowels intact while spectral shape of mid to high frequency inputs, this in turn allows
allowing compression of important high frequency information in for better recognition of speech and environmental sounds and
fricatives down to sufficiently low output frequencies. The output leads to even better spontaneous acceptance than seen with
curve shown in Figure 2 illustrates this adaptive behavior SoundRecover.
schematically.

Phonak Insight | SoundRecover2 – the adaptive frequency compression algorithm 2

Figure 3 presents the spectrograms of the sample sentence “my Evidence of Benefit
name is asa” (a) without frequency lowering, (b) with
SoundRecover and (c) with SoundRecover2 processing. Table 1 summarizes the theoretical benefits of SoundRecover2 in
Figure 3(a) shows pronounced formant structures up to 5.5 kHz at comparison to the original SoundRecover, namely the extension of
0.2 seconds to 0.5 seconds and two high frequency /s/ phonemes the fitting range and audiological benefits in terms of audibility,
at 1.2 seconds and 1.9 seconds. distinction and sound quality. Moreover, a smooth transition from
existing frequency lowering technology is expected for most users.
In particular, users of the original SoundRecover feature should be
able to switch over and acclimatize to SoundRecover2 without
concerns.

Extended fitting • fits even more severe-to-profound losses

range with acceptable sound quality
• better inclusion of left-corner and ski
slope audiograms
Improved • better audibility of high frequency sounds.
audiological See internal evidence below
Figure 3(a): Spectrogram of the sample sentence “my name is asa” without
frequency lowering.
performance • improved distinction, detection and
recognition of compressed high frequency
components
In Figure 3(b), showing SoundRecover with a cut-off frequency of
• maintains sound quality, more specifically
1500 Hz and compression ratio of 2.1, the audible bandwidth preservation of familiarity and naturalness
extends up to approximately 4000 Hz. The /s/ phonemes at 1.2 and in general and in particular for low and
1.9 seconds are compressed down into a frequency area between mid frequency components
2.5 and 4 kHz. Note that the fine-spectral structures above the cut- • better awareness of environmental sound
off frequency of 1500 Hz at the beginning of the sentence are not and as a result potentially better
fully preserved at this maximum setting. spontaneous acceptance and reduced
acclimatization time

Table 1: Theoretical benefits of SoundRecover2 relative to the original

SoundRecover.

Study results
Research has confirmed the theoretical benefits outlined in Table 1.
An early study tested the performance of 14 children with severe-
Figure 3(b): Spectrogram of the sample sentence “my name is asa” with
SoundRecover (cut-off: 1500 Hz, compression ratio: 2.1).
to-profound high frequency sensorineural hearing loss on a very
mature prototype of SoundRecover2 relative to the original
In Figure 3(c), showing SoundRecover2 with a lower cut-off SoundRecover feature (Wolfe et al. 2016). The study showed
frequency of 1479 Hz, upper cut-off frequency of 3600 Hz, and improved word recognition in quiet and improved recognition of
compression ratio of 1.4, the audible bandwidth extends up to plurals. No detriment in the detection or recognition of consonants
approximately 4000 Hz as well. Note the preservation of the was seen, and long-term users of the original SoundRecover feature
spectral fine structures up to the upper cut-off of 3600 Hz at the were able to switch to SoundRecover2 without a long period of
beginning of the sentence at 0.2 to 0.5 seconds, and the remapping acclimatization.
of the two significant high frequency /s/ phonemes at 1.2 and 1.9
seconds down into a frequency area as low as between 2000 and Another study carried out at Phonak headquarters compared the
3000 Hz. audiological performance of the original SoundRecover with
SoundRecover2 on 8 male hearing impaired adults (average age:
56.8 years) with profound symmetrical sensorineural or mixed
average loss higher than 90 dB over the frequencies from 250 Hz to
8 kHz). For this profound group, the phoneme perception test
(Schmitt et al. 2016) revealed significantly better detection
thresholds for 3 out of 4 tested stimuli (Figure 4), and a
significantly better recognition threshold for 1 out of 4 stimuli
Figure 3(c): Spectrogram of the sample sentence “my name is asa” with (Figure 5).
SoundRecover2 (lower cut-off: 1479 Hz, upper cut-off: 3600 Hz, compression
ratio: 1.4).

Phonak Insight | SoundRecover2 – the adaptive frequency compression algorithm 3

 Perceptual trade-offs in fitting frequency lowering solutions
In fitting any kind of frequency lowering scheme, a perceptual
trade-off must be managed between the balance of audiological
benefit and overall sound quality. Most contemporary frequency
lowering approaches manage this trade-off sufficiently well for
varying degrees of high frequency hearing loss. However, for some
extreme severe-to-profound hearing losses, such as those with no
aidable hearing beyond 2000 Hz, the application of frequency
lowering is a particular challenge. When frequency lowering is
applied to present high frequency sounds in the strongly limited
audible range of these patients, sound quality may be compromised
and thereby limits the benefit of frequency lowering algorithms.

This challenge is illustrated in the trade-off diagram of Figure 6

which qualitatively shows the effect of adjusting the strength of
Figure 4: Phoneme perception test: median detection thresholds for profound the frequency lowering scheme along the perceptual dimensions of
hearing losses, Naída Q SR (original SoundRecover)2 vs Naída V SR2 audibility and sound quality.
(SoundRecover2). Detection of 3 out of 4 stimuli tested (sh5, s6, s9) was
significantly better with SoundRecover2.

Figure 6: Perceptual trade-off curves for extreme severe-to-profound hearing

loss with contemporary frequency lowering solutions showing the unsatisfactory
compromise between audiological benefit and sound quality. Improved
audibility is only achieved at the expense of sound quality, and vice versa.

SoundRecover2 not only offers the technical means to overcome

the above described limitations, but also a way for hearing care
Figure 5: Phoneme perception test: median recognition thresholds for profound professionals to better manage the involved perceptual trade-offs.
hearing losses, Naída Q SR (original SoundRecover)2 vs Naída V SR2
(SoundRecover2). Recognition of 1 out of 4 stimuli tested (Asha5) was Basic principle of SoundRecover2 fitting
significantly better with SoundRecover2.
The performance of frequency lowering schemes can be
characterized in an intuitive and simple way by means of
perceptual dimensions. Thus the fitting approach for
The SoundRecover2 Fitting Concept SoundRecover2 is based on the following three important
perceptual dimensions:
Objective
From extensive research studies with patients, it is known that
 Audibility of high frequency sounds like phonemes /s/, /f/ and
achieving good audibility and distinction are among the most
/th/,
important goals for providing better hearing. At the same time,
 Distinction or discrimination of lowered high frequency
good sound quality is important for spontaneous acceptance and
sounds like /s/ and /sh/ and
for hearing comfort. The objective in fitting SoundRecover2 is to
 Sound quality of low and mid frequency sounds, like vowels
preserve, and when necessary, restore the audibility of high
/a/, /e/, /i/.
frequency sounds and the discrimination of compressed high
frequency sounds, and to maintain the familiarity of the entire
These three interrelated perceptual dimensions can be depicted
frequency range.
graphically with the “trade-off triangle”. Figure 7 illustrates three
possible configurations of the trade-off triangle, a default setting, a
weaker setting and a stronger setting.
2
Where detection or recognition thresholds were not measureable with
the original SoundRecover (s9, Asa6 and Asa9 stimuli), the thresholds
were set to 75 dB.

Phonak Insight | SoundRecover2 – the adaptive frequency compression algorithm 4

 adjustment towards better distinction results in a change of
compression ratio. The third dimension, sound quality, is
automatically optimized for each change of the essential fitting
dimensions within the fitting software. Thus, when the balance
between audibility and distinction has been optimized for a
particular user, sound quality is always automatically adjusted to
maintain optimal clarity.

Sound quality can be characterized in more detail in terms of sound

clarity and hearing comfort. On occasion, a patient may need a
different balance between clarity and comfort. In this case, it is
possible to further adjust the sound quality with 4 predefined
settings to balance optimum clarity and personal hearing comfort.
An adjustment to increase hearing comfort results in a shift of the
Figure 7: The SoundRecover2 trade-off triangle for a default SoundRecover2
setting (green), a setting with stronger audibility (blue) and another setting
upper cut-off frequency towards the maximum audible bandwidth.
with stronger distinction (purple). Note the effects of modifications on one
perceptual dimension onto the other two dimensions. Figure 8 illustrates the fitting procedure for SoundRecover2 with
fine-tuning of audibility and distinction and subsequent fine-
SoundRecover2 aims to optimize the balance between these tuning of sound quality. This method is implemented with two
interdependent perceptual dimensions by: perceptual sliders within the fitting software.

1) making use of the individual’s audible bandwidth in the

best possible way
 the maximum output frequency is set to the upper limit of
the individual’s audible bandwidth in order to maximize
stimulation of the hearing nerve where audibility can be
achieved without frequency lowering
 no risk of deprivation
2) protecting the mid and low frequencies in the best possible
way
 the upper cut-off frequency is set high enough so that
speech which is audible without compression remains
unaffected
 no risk for distortion of mid and low frequencies
3) presenting the compressed sounds in an optimally selected Figure 8: The SoundRecover2 fitting procedure. The basis of the fitting is the
frequency area default setting as derived from pre-calculation. Fine-tuning allows the
optimization of the audibility of high frequency sounds and the distinction of
 the starting point of compression (the lower cut-off) is set the lowered high frequency components. The resulting settings include an
at the lowest possible frequency yielding a wide area of automatically optimized sound quality to achieve maximum sound clarity. The
compression with weak compression ratio sound quality can be further fine-tuned towards more hearing comfort in four
predefined steps.
 increases the overall strength of frequency compression for
those extreme severe-to-profound losses extending the
Outcomes
benefits of frequency lowering to more users.
This fitting procedure, based on perceptual dimensions, allows the
hearing care professional to adjust SoundRecover2 in an intuitive,
Pre-calculation
understandable and user-friendly way. Reports from patients during
Taking into account these three elements and based on subject
frequency lowering fitting relate much better to perceptual
data collected over several years, an optimized pre-calculation and
dimensions than to technical parameters or the “weaker-stronger”
starting point for fitting has been carefully developed. The resulting
control of the original SoundRecover.
default setting for all fittings is the best possible balance across the
three perceptual dimensions, in particular, allowing for good
The default settings derived from the pre-calculation set an
audibility of high frequency sounds while enabling sufficient
excellent starting point for the fitting by providing good audibility,
distinction between compressed sounds, and at the same time
together with sufficient distinction of compressed sounds and
yielding an acceptable overall sound quality.
acceptable sound quality.
SoundRecover2 fitting procedure
Fine-tuning allows for individual adjustment, if required, of the two
dimensions audibility and distinction. An adjustment for greater
audibility results in a decreased lower cut-off frequency, while an

Phonak Insight | SoundRecover2 – the adaptive frequency compression algorithm 5

Conclusions Authors
The new SoundRecover2 frequency lowering algorithm is Julia Rehmann is an Audiological Engineer at Sonova Headquarters
specifically designed to allow operation with lower cut-off in Switzerland.
frequencies and weaker compression ratios in order to expand the
already existing user base of SoundRecover. This ambitious goal Siddhartha Jha is a DSP Engineer in the Science & Technology
was accomplished by introducing an additional cut-off frequency department at Sonova Headquarters in Switzerland.
and by adaptively adjusting the starting point of compression to
the energy content of the input signal. By instantaneously Silvia Allegro Baumann is a Senior DSP Engineer in the Science &
switching between the lower and upper cut-off frequencies, high Technology department at Sonova Headquarters in Switzerland.
frequency components can be compressed into a larger frequency
range with a weaker compression ratio, while low frequency
components can be protected and are left unaltered.

A preliminary external study showed improved word recognition in

quiet and improved recognition of plurals, with no detriment in the
detection or recognition of consonants. Long-term users of the
original SoundRecover feature were able to switch to
SoundRecover2 without a long period of acclimatization (Wolfe et
al. 2016). In an internal study at Phonak headquarters with a group
of subjects with profound hearing loss, the Phoneme Perception
Test revealed significantly better detection thresholds for 3 out of 4
tested stimuli and a significantly better recognition threshold for 1
out of 4 high frequency stimuli.

Together with the new signal processing, a new fitting concept has
been developed which allows easier management of the difficult
trade-off between audiological benefits and sound quality in
frequency lowering systems. The resulting perceptual fitting is
based on the “trade-off triangle” which depicts the interrelated
perceptual dimensions: audibility, distinction and sound quality.
The fitting pre-calculation offers an optimal starting point for good
audibility of high frequency sounds, enables sufficient distinction
between compressed sounds and delivers an acceptable overall
sound quality. Fine-tuning to match users’ individual needs can
easily be carried out with the use of perceptual sliders in the Target
fitting software.

References
028-1512-02/V2.00/2016-01/ © Phonak AG All rights reserved

McDermott, H. (2010). SoundRecover - The importance of wide

perceptual bandwidth. Phonak Background Story.

Wolfe, J., Duke, M., Schafer, E., Rehmann, J., Jha, S., John, A., Jones,
C. (2016). Preliminary evaluation of a novel non-linear frequency
compression scheme for use in children. Submitted to International
Journal of Audiology.

Schmitt, N., Winkler, A., Boretzki, M., Holube, I. (2016). A phoneme

perception test method for high frequency hearing aid fitting.
Journal of the American Academy of Audiology.

Phonak Insight | SoundRecover2 – the adaptive frequency compression algorithm 6