Приложение Petralex для обучения слабослышащих студентов

Implementation of a Digital Hearing Aid with
User-Settable Frequency Response and Sliding-Band
Dynamic Range Compression as a Smartphone App
Saketh Sharma, Nitya Tiwari, and Prem C. Pandey
Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai, India
{sakethsharma,nitya,pcpandey}@ee.iitb.ac.in
Abstract. Persons with sensorineural hearing loss suffer from degraded speech
perception caused by frequency-dependent elevation of hearing thresholds, re-
duced dynamic range and abnormal loudness growth, and increased temporal and
spectral masking. For improving speech perception by persons with moderate
loss of this type, a digital hearing aid is implemented as an Android-based
smartphone app. It uses sliding-band dynamic range compression for restoring
normal loudness of low-level sounds without making the high-level sounds un-
comfortably loud and for reducing the perceptible temporal and spectral distor-
tions associated with currently used single and multiband compression tech-
niques. The processing involves application of a frequency dependent gain func-
tion calculated on the basis of critical bandwidth based short-time power spec-
trum and is realized using FFT-based analysis-synthesis. The implementation has
a touch-controlled graphical user interface enabling the app user to fine tune the
frequency-dependent parameters in an interactive and real-time mode.
Keywords. Dynamic range compression ∙ Hearing aid ∙ Interactive settings ∙ Sen-
sorineural hearing loss ∙ Smartphone app.
1 Introduction
Abnormalities in the sensory hair cells or the auditory nerve result in sensorineural
hearing loss. It is a commonly occurring hearing loss, which may be inherited genet-
ically or may be caused by excessive noise exposure, aging, infection, or use of ototoxic
drugs. It is associated with frequency-dependent elevation of hearing thresholds, re-
duced dynamic range along with abnormal growth of loudness known as loudness re-
cruitment, and increased temporal and spectral masking leading to degraded speech
perception [1, 2, 3]. Hearing aids used for alleviating the effects of sensorineural hear-
ing loss provide frequency-selective amplification along with dynamic range compres-
sion [4, 5, 6, 7, 8]. They may have directional microphones for improving the signal-
to-noise ratio and may employ signal processing for suppression of wind and back-
ground noise, acoustic feedback cancellation, environment learning for selection of the
frequency response, and generation of comfort noise for patients with tinnitus. Some
aids also provide facility of customization using mobile apps and wireless connectivity.
Frequency-dependent elevation of hearing thresholds occurs due to loss of inner hair
cells and leads to reduced audibility of speech. Loss of outer hair cells results in reduced
Int. Conf. on Intelligent Human Computer Interaction, 12-13 December, 2016, Pilani, Raj., India (IHCI 2016), Paper No. 81
1

dynamic range, i.e. increase in hearing threshold levels without corresponding increase
in upper comfortable levels. It also distorts the loudness relationships among speech
components leading to degraded speech perception [2, 3]. As linear amplification of
sounds is unsuitable for persons with sensorineural hearing loss, hearing aids provide
amplification along with dynamic range compression to compensate for reduced dy-
namic range. The gains and compressions are usually set by an audiologist based on the
patient's audiogram and in accordance with a hearing aid fitting protocol [4, 6, 7].
Dynamic range compression involves changing the gain based on the input signal
level for presenting the sounds comfortably within the limited dynamic range of the
hearing impaired listener [4, 5, 8, 9, 10, 11]. The processing is carried out as single-
band compression or as multiband compression. In single band compression, the gain
is calculated as a function of the signal power over its entire bandwidth. As speech
signal is dominated by low frequency components, this compression distorts the per-
ceived temporal envelope of the signal and often makes the high frequency components
inaudible. Most of the currently available hearing aids provide multiband dynamic
range compression which involves dividing the input signal in multiple bands and ap-
plying a gain in each band which is a function of the signal power in that band. It avoids
the problems associated with single band compression, but results in decreased spectral
contrasts and modulation depths in the speech signal. Further, different gains in adja-
cent bands may distort spectral shape of a formant spanning the band boundaries, par-
ticularly during formant transitions. With the objective of reducing the temporal and
spectral distortions associated with the compression techniques currently used in the
hearing aids, a sliding-band compression technique has been reported by Tiwari and
Pandey [12, 13] for use in digital hearing aids. It calculates a frequency-dependent gain
function, wherein the gain for each frequency sample is calculated as a function of the
signal power in the auditory critical band centered at it. It keeps compression related
distortions below perceptible levels for improving speech perception.
Hearing aids are designed using ASICs due to power and size constraints. Therefore,
incorporation of a new compression technique in hearing aids and its field evaluation
is prohibitively expensive. Use of smartphone-based application software (app) to cus-
tomize and remotely configure settings on hearing aids provide a greater flexibility to
hearing aid users and developers. Many hearing aid manufacturers (GN ReSound, Pho-
nak, Unitron, Siemens, etc) provide apps to control hearing aids using Android or iOS
smartphone. This type of app helps the hearing aid user in personalizing the listening
experience by adjustment of settings during use of the device and avoids repeated visits
to an audiology clinic. The smartphone-based apps may also be used for development
and testing of signal processing techniques for hearing aids. Hearing aid apps [15, 16,
17, 18, 19, 20] such as 'Petralex', 'uSound', 'Q+
', and 'BioAid' for Android/iOS, ‘Mimi’,
'EnhancedEars' for iOS, and "Hearing Aid with Replay" and “Ear Assist” for Android
provide users with moderate sensorineural hearing loss a low-cost alternative for hear-
ing aids. In addition to providing frequency-selective gain and multiband dynamic
range compression, they also offer the flexibility of creating and storing sound profiles
specific to the user's hearing loss characteristics.
Implementation of the sliding-band compression technique for real-time processing
using a 16-bit fixed-point DSP chip with on-chip FFT hardware has been reported in
2

[12] and its implementation as an iPhone-based hearing aid app has been reported in
[14]. However, these implementations use predefined processing parameters. To make
this low-distortion compression technique conveniently available to the hearing im-
paired persons and to permit its evaluation by a large number of users without incurring
the expenses involved in the ASIC-based hearing aid development, a smartphone app
implementing the digital hearing aid with user settable frequency response and dynamic
range compression is needed. Such an app is developed for providing the user flexibility
to fine tune the frequency-dependent parameters in an interactive and real-time mode by
a touch-controlled graphical user interface (GUI). The implementation is carried out us-
ing an Android-based smartphone handset, as such handsets are available from a number
of manufacturers with features and cost suiting a large number of users. The handset
model 'Nexus 5X’ with Android 7.0 is used for implementation and testing on the basis
of its relatively low audio I/O (input-output) delay and high processing capacity. Most
handset models may meet these criteria in the near future.
The signal processing for the app is described in the second section. The third section
presents the details of its implementation as a smartphone app. Test results are pre-
sented in the fourth section, followed by conclusions in the last section.
2 Signal Processing for Settable Frequency Response and
Sliding-Band Dynamic Range Compression
Block diagram of the signal processing for the app is shown in Fig. 1. The operations
for compensation of increased hearing thresholds and decreased dynamic range are car-
ried out by spectral modification. The processing uses a DFT-based analysis-synthesis
comprising the steps of short-time spectral analysis, frequency and level dependent
spectral modification, and signal resynthesis. The analysis involves segmentation of the
input signal into overlapping frames and calculating DFT to get short-time complex
spectra. A frequency-dependent gain function is calculated in accordance with the con-
trol settings and as a function of the short-time magnitude spectrum. This gain function
is used for modification of the short-time input complex spectrum and its IDFT is cal-
culated to get the overlapping output signal frames. The output signal is re-synthesized
using an overlap-add technique for reducing the processing related artifacts.
In our design, it is possible to use a look-up table to provide the gain function most
appropriate for improving speech perception by the listener. However, assessment of
the loudness growth function as a function of frequency to derive the look-up table is
not practical. Hence, the control settings are in the form of plots of desired levels for
‘soft’, ‘comfortable’, and ‘loud’ sounds (referred to as SL, CL, LL, respectively) and
Fig. 1. Signal processing in the hearing aid app with settable frequency response and dynamic
range compression.
3

input using touch-controlled GUI in an interactive manner. These values are used for
calculating the frequency and level dependent gain function. For each frequency sample
k, the spectral modification is carried out using a piece-wise linear relation between the
input power and the output power on dB scale as shown in Fig. 2. The relationship is
specified by the values of , , and which are the output signal
levels corresponding to soft, comfortable, and loud sounds, respectively, for the hearing
aid user and by the values of and which are the input signal levels
corresponding to soft and loud sounds, respectively, for a normal-hearing listener.
The lower segment of the input-output relationship, marked as ‘CR = 1’ in Fig. 2,
corresponds to linear gain providing the amplification needed for soft level sounds and
the value of the gain is given as
(1)
This gain is applied unless the output exceeds the comfortable level. Thus the input-
output relationship in this segment for ith frame is given as the following:
(2)
The central segment of the relationship, marked as ‘CR > 1’, corresponds to com-
pression with compression ratio given as
(3)
This relationship is applicable for the output lying between comfortable and loud levels,
and is given as the following:
(4)
The upper segment of the relationship, marked as ‘CR = ∞’, corresponds to limiting of
the output power and is applied when the output exceeds loud level. The relationship is
given as the following:
Fig. 2. Relation between input power (dB) and output power (dB) for ith frame and band centered
at kth spectral sample.
4

(5)
In accordance with the input-output relationship as represented by the three seg-
ments and given in Eqs. 2, 4, and 5, the target gain for the spectral sample k in ith frame
is given as:
(6)
Block diagram of the spectral modification is shown in Fig. 3. For each frequency
sample k in ith frame, the input level PIdB(i,k) is calculated as the sum of squared mag-
nitude of the spectral samples in the band centered at k and with bandwidth correspond-
ing to auditory critical bandwidth given as
(7)
where f(k) is the frequency of the kth spectral sample in kHz.
For spectral modification, the target gain is converted to linear scale. The gain ap-
plied to the kth spectral sample in the ith frame is obtained using the desired attack and
release times by updating the gain from the previous value towards the target value, as
given in Eq. 6, and is given as
(8)
The number of steps during the attack and release phases are controlled using gain ratios
and , respectively. Here and are
the maximum and minimum values of target gain. The number of steps during attack
and the number of steps during release are selected to set the attack time as
Fig. 3. Spectral modification for compensation of increased hearing thresholds and decreased
dynamic range using sliding-band dynamic range compression.
5

and release time as where is the sampling frequency and is the
number of samples for frame shift. A fast attack avoids the output level from exceeding
the uncomfortable level during transients, and a slow release avoids the pumping effect
or amplification of breathing.
The processing for dynamic range compression modifies the magnitude spectrum
and the signal is resynthesized using the original phase spectrum. It results in percepti-
ble distortions due to phase discontinuities in the modified short-time complex spec-
trum. To mask these distortions, the analysis-synthesis method based on the least
squared error estimation (LSEE) as proposed by Griffin and Lim [21] is used. It in-
volves segmenting the input signal using L-sample frames with 75% overlap, i.e. frame
shift , and multiplying the segmented frames with modified-Hamming window.
Complex spectrum of -sample frame is obtained after zero-padding it to length and
calculating -point FFT. After spectral modification as described earlier, -point IFFT
is calculated to get the modified output frame which is multiplied with the modified-
Hamming window. The successive output frames are added in accordance with the
overlap of the input frames to provide the resynthesized output signal.
3 App Implementation for Real-Time Processing
A signal bandwidth of 4 kHz is generally considered adequate for hearing aid appli-
cation, and hence a sampling frequency of 10 kHz may be used. As the detectability
threshold for audio-visual delay is reported to be 125 ms [22], the signal delay intro-
duced by the app should be less than this value to avoid audio-visual de-synchrony
during face-to-face communication. After examining the processing capability and au-
dio I/O delay of several currently available Android-based handsets [23], implementa-
tion of the app for real-time processing has been carried out and tested using ‘Nexus
5X’ with Android 7.0 Nougat OS. It provides a default audio sampling frequency of 48
kHz which can be decreased using the re-sampler block but with an increase in the I/O
delay due to the antialiasing and smoothening filters. Considering the computational
load associated with sampling frequency of 48 kHz and increased delay with sampling
frequency of 12 kHz, the processing is carried out with a down-sampling ratio of two
resulting in sampling frequency of 24 kHz. The FFT-based analysis-synthesis uses 20-
ms frames with 75% frame overlap and 1024-point FFT. Thus the implementation uses
fs = 24 kHz, L=480, S=120, and N=1024.
The app “Hearing Assistant 1.0” has been developed with a GUI for control settings,
enabling graphical setting of the controls for threshold and range compensations. The
screen is used in its landscape mode to make an appropriate use of the aspect-ratio,
ignoring the ‘auto-rotate’ setting. The screenshot of the home screen of the app is shown
in Fig. 4. The play/stop button is for on/off control. The two bars at the right side graph-
ically display the input and output levels for assessing the effect of control settings. The
graphical control for gain and range compensation is accessed through the ‘settings’
button of ‘compression’ module. A provision has been provided to include other pro-
cessing modules, such as noise suppression [24, 25] and CVR enhancement [26] in the
later versions of the app. All modules have on/off and ‘settings’ buttons. The on/off
6

button can be used for toggling the function and the settings button can be used for
setting the processing parameters graphically.
Fig. 5 shows the screenshot of the ‘settings’ screen for compression. It provides
touch control of points, called ‘thumbs’, on the level vs frequency plots for setting the
output levels for soft, comfortable, and loud sounds, for up to 10 frequencies. A thumb
can be moved by dragging, freely in the horizontal direction and as constrained by the
upper and lower curves in the vertical direction. Movement of a thumb on one curve
along x-axis moves the corresponding thumbs on the other two curves, while the move-
ment along y-axis does not affect them. When a touch event occurs, the thumbs on the
three curves closer with respect to the x-position of the touch are shifted to the new x-
position, and the closest one with respect to the y-position is shifted to the new y-posi-
tion. This method permits changing a thumb position without dragging it from its earlier
position. With each change in the thumb position, the curves for all the intermediate
frequency values are obtained by a smooth curve fitting through the thumbs, using
Fritsch-Carlson method of monotonic cubic spline fitting [27] to ensure that the curve
Fig. 4. Screenshot of home screen of hearing aid app “Hearing Assistant 1.0”
Fig. 5. Screenshot of settings screen to set parameters of frequency dependent gain and dynamic
range compression
7

between two adjacent control points is monotonic. The gains calculated are transferred
to the spectral modification block at the next analysis frame, permitting a real-time and
interactive touch-control of the processing parameters. The values of input and output
levels in each frame are transferred to GUI for updating the level bars.
The program was written using a combination of C++ and Java, with Android Studio
2.1.0 as the development environment. The block diagram of application process of the
hearing aid app is shown in Fig. 6. As the audio I/O delay using Java APIs (application
programming interfaces) of Android is beyond the acceptable value of 125 ms, the I/O
framework is implemented in C++ using Open SLES library [28] which is a hardware-
accelerated audio API for input/output handling. An Open SLES audio engine with
recording and playback capabilities is realized. Since the application is required to run
even when other applications are switched on, the audio engine is maintained using a
high priority background service. The service is started when the application comes to
foreground and remains running until the playback is stopped or the application process
is killed. The communication between GUI and the audio engine is carried out through
background service using JNI (Java native interface) as shown in Fig. 6.
A recorder and a player object are realized with required sampling rate and frame
length specifications, each with internal queues to manage the buffer blocks currently
being recorded/played. Separate threads are spawned for recorder and player to ensure
uninterrupted I/O. The recorder and player threads are scheduled at least once within
the frame shift duration of 5 ms. Blocking calls to the operating system (OS) such as
memory allocation, mutex locks, etc. can cause underruns in real-time applications. To
minimize the blocking calls, the memory required to handle I/O and processing are
allocated while creating the audio engine. The processing and the signal acquisition and
playback are carried out using pointers and no memory allocation is required while the
app is running. The new samples from ADC are filled in the S-word free buffer of
recorder thread. The recorder thread uses 2S-word buffer queue and it passes the pointer
of S-word filled buffer to JNI interface when scheduled. The processing thread when
scheduled gets pointer of S-word buffer containing the new samples from the JNI in-
terface and returns the pointer of S-word buffer containing processed samples. The
player buffer when scheduled gets pointer of S-word buffer containing processed sam-
ples from JNI interface and these samples are output through DAC.
In the processing thread, the input samples, real and imaginary parts of spectral
samples, and the output samples are stored as 32-bit floating point arrays. The input
samples are acquired in S-word blocks from JNI interface. For each S-word just-filled
Fig. 6. Block diagram of application process in the hearing aid app.
8

block, a 3S-word buffer stores the previous input samples. Input window with L samples
(4S words) is formed using the samples of the just-filled block and the samples from
previous three blocks stored in 3S-word buffer. These L samples are multiplied by L-
point modified-Hamming window and N-point FFT is calculated. Open source Kiss FFT
library [29] has been used for floating point FFT computation. Input power in an auditory
critical band centered at each of the first N/2 complex spectral components is calculated
as given in Eq. 7. Modified spectrum is obtained by multiplying first N/2 complex spec-
tral samples with the corresponding gains. The other samples of modified spectrum are
obtained by taking complex conjugate of first N/2 complex spectral samples. N-point
IFFT of the modified complex spectrum is calculated and the resulting sequence is mul-
tiplied with the modified-Hamming window to get the modified output frame. The out-
put signal is re-synthesized using overlap-add. In the current implementation, the pro-
cessed signal is output on both the audio channels. Separate graphical settings of the
processing parameters for the two channels may be incorporated for use of the app as a
binaural hearing aid. The processing has algorithmic delay of one frame, i.e. 20 ms. For
real-time output, the processing of each frame has to get completed before the arrival of
the next frame, i.e. the frame shift interval of 5 ms, resulting in the signal processing
delay, comprising the algorithmic and computational delays, of 20 – 25 ms.
4 Test Results
The app was tested on the handset model ‘Nexus 5X’ with Android 7.0 OS. Quali-
tative evaluation was carried out using the headset of the handset for speech input
through its microphone and audio output through its earphone. The objective evaluation
of the processing was carried out using an audio input-output interface with a 4-pin
TRRS (tip-ring-ring-sleeve) connector to the headset port of the handset, as shown in
Fig. 7. It has a resistive attenuator for attenuating the input audio signal to a level com-
patible with the microphone signal level and output resistance of 1.8 k for it to be
recognized as external microphone. The two output channels have 100 load re-
sistances. A PC sound card was used for applying the audio input and acquiring the
processed output. Additional testing was carried out by providing signals from a func-
tion generator and observing the waveform using a digital storage oscilloscope.
The settable frequency response without compression was tested using a sinusoidal
input of swept-frequency and constant amplitude. An example of the processing along
with the frequency-dependent gains set by the GUI is shown in Fig. 8, with the soft
level kept high in bands at 1 kHz and 7.5 kHz, and the comfortable level and the loud
Fig. 7. Audio interface to the 4-pin TRRS headset port of the mobile handset
9

level kept same and at the maximum value. The input frequency is linearly swept from
100 Hz to 10 kHz over 10 s. The processed output shows changes in amplitude of the
sine wave in accordance with the soft-level curve. The processing for dynamic range
compression was tested by setting different values for comfortable and loud levels and
amplitude-modulated tones of different frequency and amplitude envelopes as applied
inputs. An example of the compression is shown in Fig. 9. The input is an amplitude-
modulated tone of 1 kHz. The processing gives higher gains at lower values of the input
level. Spikes in the amplitude envelope of the output signal in response to step changes
in the amplitude envelope of the input signal, as seen in the figure, are typical of the
dynamic range compression with a finite frame shift and can be eliminated by using
one-sample frame shift but with a significantly increased computation load.
The app was tested for speech modulated with different types of amplitude enve-
lopes. An example of the processing is shown in Fig. 10, with an English sentence
repeatedly concatenated with different scaling factors to observe the effect of variation
in the input level on the output waveform. Informal listening test was carried out with
different speech materials, music, and environmental sounds with large variation in the
sound level as inputs. The outputs exhibited the desired amplification and compression
without introducing perceptible distortions. There were no perceptible differences be-
tween the processed outputs from the app and the offline implementation. Examples of
the processing by the app are available at [30].
The audio latency, i.e. the signal delay comprising the processing delay and I/O
delay, was measured by applying a 1 kHz tone burst of 200 ms from a function gener-
ator as the input and observing the delay from onset of the input tone burst to the
Fig. 8. Example of processing for frequency-dependent linear gain: (a) parameters, (b) input of
tone of constant amplitude with frequency linearly swept from 100 Hz to 10 kHz over 10 s, and
(c) processed output
(a)
(b)
(c)
10

corresponding onset in the output, using a digital storage oscilloscope. It was found to
be approximately 55 ms. The processing delay (sum of the algorithmic and computa-
tional delays) is 1.25 times the frame length, i.e. approximately 25 ms. The additional
Fig. 9. Example of processing for compression: (a) parameters, (b) input of amplitude-modulated
sinusoidal tone of 1 kHz, and (c) processed output
(a)
(b)
(c)
(a)
(b)
(c)
Fig. 10. Example of processing of speech signal: (a) parameters, (b) input speech signal with
amplitude modulation (English sentence "you will mark ut please" repeatedly concatenated with
different scaling factors), and (c) processed output
11

delay is due to audio I/O delay of the handset hardware, buffering operations in the OS,
and delays in the anti-aliasing and smoothening filters. The audio latency of our imple-
mentation is much lower than the detectability threshold of 125 ms and hence it should
be acceptable for face-to-face communication.
5 Conclusion
A smartphone app has been developed for implementing a digital hearing aid incor-
porating user settable frequency response and sliding-band dynamic range compression
for improving speech perception by persons with moderate sensorineural hearing loss. It
is aimed at enabling the hearing of soft-to-loud speech without introducing perceptible
distortions. The processing parameters can be set by the user in an interactive and real-
time mode using a graphical touch interface. The implementation has been validated by
graphical assessment of the processed outputs and informal listening tests. The audio
latency of the app is acceptable for audio-visual perception permitting its use as a hearing
aid during face-to-face conversation. Use of the app by a large number of hearing im-
paired listeners is needed for its real life evaluation and further enhancement. Interactive
settings and processing for binaural hearing aid may be incorporated in the app. It may
be extended as a hearing aid for persons with severe loss by using headsets with appro-
priate output levels. With further increase in the processing capacity and decrease in
audio I/O delay of newer handset models, there is scope for incorporating processing for
speech enhancement in the app for extending its usability in noisy environments.
Acknowledgment
The research is supported by “National Program on Perception Engineering Phase-
II” sponsored by the Department of Electronics & Information Technology, Govern-
ment of India.
References
1. Levitt, H., Pickett, J.M., Houde, R.A. (eds.): Sensory Aids for the Hearing Impaired. IEEE Press, New
York (1980)
2. Moore, B.C.J.: An Introduction to the Psychology of Hearing. Academic, London, UK, pp 66–107
(1997).
3. Gelfand, S.A.: Hearing: An Introduction to Psychological and Physiological Acoustics. 3rd ed., Marcel
Dekker, New York, pp. 314–318 (1998)
4. Dillon, H.: Hearing Aids. Thieme Medical, New York (2001)
5. Sandlin, R.E.: Textbook of Hearing Aid Amplification. Singular, San Diego, Cal., pp. 210–220 (2000)
6. Byrne, D., Tonnison, W.: Selecting the gain of hearing aids for persons with sensorineural hearing im-
pairments. Scand. Audiol. 5, 51–59 (1976)
7. McCandless, G.A., Lyregaard, P.E.: Prescription of gain/output (POGO) for hearing aids. Hear Instrum.
34, 1, 16–21 (1983)
8. Braida, L.D., Durlach, N.I., Lippmann, R.P., Hicks, B.L., Rabinowitz, W.M., Reed, C.M.: Hearing aids
– A review of past research on linear amplification, amplitude compression, and frequency lowering.
12

Hardy, J.C. (ed.) ASHA Monographs, Rockville, MA (1979) www.asha.org/uploadedFiles/publications/
archive/Monographs19.pdf
9. Lippmann, R.P., Braida, L.D., Durlach, N.I.: Study of multichannel amplitude compression and linear
amplification for persons with sensorineural hearing loss. J. Acoust. Soc. Am. 69, 524–534 (1981)
10. Asano, F., Suzuki, Y., Sone, T., Kakehata, S., Satake, M., Ohyama, K., Kobayashi, T., Takasaka, T.: A
digital hearing aid that compensates for sensorineural impaired listeners. In: Proc. IEEE Int. Conf.
Acoust. Speech Signal Process. 1991 (ICASSP 1991), Toronto, Canada, pp. 3625–3628
11. Stone, M.A., Moore, B.C.J., Alcántara, J.I., Glasberg, B.R.: Comparison of different forms of compres-
sion using wearable digital hearing aids. J. Acoust. Soc. Am. 106, 3603–3619 (1999)
12. Tiwari, N., Pandey, P.C.: A sliding-band dynamic range compression for use in hearing aids. In: Proc.
20th Nat. Conf. Commun. (NCC 2014), Kanpur, India (2014) doi: 10.1109/NCC.2014.6811300.
13. Pandey, P.C., Tiwari, N.: Dynamic range compression with low distortion for use in hearing aids and
audio systems, International PCT Application number PCT/IN2015/000049 (2015)
14. Tiwari, N., Pandey, P.C., Sharma A.: A smartphone app-based digital hearing aid with sliding-band dy-
namic range compression. In: Proc. 22nd Nat. Conf. Commun. (NCC 2016), Guwahati, India (2016) doi:
10.1109/NCC.2016.7561146
15. Mimi Hearing Technologies: Mimi hearing test v3.0. itunes.apple.com/in/app/mimi-hearing-
test/id932496645
16. IT4You: Petralex hearing aid v1.4.3. play.google.com/store/apps/details?id=com.it4you.petralex,
http://itunes.apple.com/us/app/petralex-hearing-id
17. Clark, N.: Bio aid v1.0.2. itunes.apple.com/gb/app/bioaid
18. Lamberg Solutions: Hearing aid with replay v2.0.0. play.google.com/store/apps/details?id=com.
ls.sound amplifier
19. Newbrick, S.A.: uSound (hearing assistant) v.1.6. play.google.com/store/apps/details?id=com.
newbrick.usound
20. Quadio Devices.: Q+ hearing aid. play.google.com/store/apps/details?id=com.quadiodevices.qplus
21. Griffin, D.W., Lim, J.S.: Signal estimation from modified short-time Fourier transform. IEEE Trans.
Acoust. Speech Signal Process. 32, 236 – 243 (1984)
22. International Telecommunication Union: Relative timing of sound and vision for broadcasting, ITU Rec.
ITU-R BT.1359 (1998)
23. Superpowered Inc: Android audio latency and iOS audio latency test app superpowered.com/latency
24. Loizou, P.C.: Speech Enhancement: Theory and Practice, CRC, New York (2007)
25. Tiwari, N., Pandey, P.C.: Speech enhancement using noise estimation based on dynamic quantile track-
ing for hearing impaired listeners. In: Proc. 21st Nat. Conf. Commun. (NCC 2015), Mumbai, India
(2015) doi: 10.1109/NCC.2015.7084849
26. Jayan, A.R., Pandey, P.C.: Automated modification of consonant-vowel ratio of stops for improving
speech intelligibility, Int. J. Speech Technol., 18, 113-130 (2015)
27. Fritsch, F.N., Carlson, R.E.: Monotone piecewise cubic interpolation, SIAM J. Numer. Anal., 17, 2,
238–246 (1980)
28. Khronos group: OpenSLES. www.khronos.org/opensles/
29. Borgerging, M.: Kiss FFT. sourceforge.net/projects/kissfft
30. Sharma, S.: Dynamic range compression results from smartphone app implementation (2016)
www.ee.iitb.ac.in/~spilab/ material/saketh/ihci2016
13

Приложение Petralex для обучения слабослышащих студентов

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to Приложение Petralex для обучения слабослышащих студентов (20)

More from Danil Dintsis, Ph. D., PgMP (16)

Recently uploaded (20)

Приложение Petralex для обучения слабослышащих студентов