Sehr geehrte Abonnenten des Kolloquium-Newsletters,
gerne informieren wir Sie über den nächsten Termin unseres
Kommunikationstechnischen Kolloquiums.
*Dienstag, 6. April 2021*
*Vortragender*: Florian Hilgemann
*Zeit*: 10:00 Uhr
*Ort*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master-Vortrag*: Numerische Optimierung rekursiver Filter für aktive
akustische Equalisierung
Aktive Störgeräuschunterdrückung (ANC) ist eine auf dem schnell
wachsenden Markt für Hörsysteme von Kunden häufig nachgefragte Funktion
und hat in der Vergangenheit zunehmend an Bedeutung gewonnen. Trotz
jüngster Fortschritte im Bereich der digitalen Signalprozessoren (DSPs)
werden ANC-Systeme zwangs hoher Anforderungen an Gehäusegröße und
Stromverbrauch zusammen mit der Notwendigkeit hoher Abtastraten häufig
mittels zeitinvarianten Digitalfiltern realisiert. In diesem Fall bieten
Filter mit unendlicher Impulsantwort (IIR-Filter) eine attraktive Option
für die Implementierung, da sie eine geringere Anzahl an Multiplizierern
als Filter mit endlicher Impulsantwort (FIR-Filter) benötigen um eine
Impulsantwort mit einer bestimmten Länge zu modellieren. Leider liefern
herkömmliche Methoden für den Entwurf zeitinvarianter ANC-Filter
entweder ein FIR-Filter oder erfordern andere Formen der
Nachbearbeitung. Dies kann das Systemverhalten nachteilig beeinflussen
und im Falle von Feedback ANC kann die Stabilität des Systems nicht ohne
weiteres garantiert werden.
Um die Notwendigkeit der Nachbearbeitung zu überwinden wird in dieser
Arbeit ein alternativer Ansatz für den Entwurf von IIR-Filtern für
Feedforward und Feedback ANC untersucht, welcher auf iterativen
Optimierungsalgorithmen basiert. Dies führt unweigerlich zu
nichtkonvexen Optimierungsproblemen, deren globales Optimum nicht ohne
weiteres gefunden werden kann. Es wird jedoch gezeigt, dass der Ansatz
trotz dieser Einschränkung eine praktikable Option darstellt. Eine
wesentliche Neuerung des erarbeiteten Konzepts ist ein modularer Ansatz,
der eine präzise Formulierung des Optimierungsziels unter
Nebenbedingungen sowie eine flexible Anpassung an geänderte
Anforderungen ermöglicht. Der Ansatz wird sowohl durch Simulationen als
auch durch eine Echtzeitmessung mit einem Kunstkopf verifiziert.
Weiterhin werden die Ergebnisse in Bezug zu einem theoretischen Optimum
gesetzt, welches als Abschätzung für die mit einem zeitinvarianten
Filter erzielbare Funktionsfähigkeit dient. Insgesamt deuten die
Ergebnisse darauf hin, dass die resultierenden IIR-Filter gleich gut
oder in vielen Fällen sogar besser als konventionelle Methoden
abschneiden. Damit stellt der erarbeitete Ansatz einen wesentlichen
Schritt in Richtung optimaler Funktionsfähigkeit dar.
Alle Interessierten sich herzlich eingeladen, eine Anmeldung ist nicht
erforderlich.
Allgemeine Informationen zum Kolloquium sowie eine aktuelle Liste der
Termine des Kommunikationstechnischen Kolloquiums finden Sie unter:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Wednesday, January 20, 2021*
*Speaker:* Jonas Kuntzer
*Time*: 11:00 a.m.
*Location:*
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master Lecture:* Speech Inpainting Using Recurrent Neural Networks
Speech signals are often subject to transient noise signals that can
appear anywhere in the frequency spectrum. When the noise over laps with
frequencies of human speech, the removal of the noise inevitably leads
to a deterioration in the quality of speech and a loss of information.
Because speech is highly predictable in time, a speech inpainting neural
network is presented which includes recurrent neural networks in the
form of LSTMs whose strength lies in the prediction of sequential data.
Furthermore, the SHAP algorithm is employed to gain an understanding of
the impact of individual input features on the output and the network's
dependence on individual time steps in the input sequences.
The results demonstrate that LSTMs are well-suited to solve the problem
of speech inpainting, outperforming other networks based on
fully-connceted layers only. An increase in the PESQ and LSD score is
observed compared to the corrupted signal. The increase is further
enhanced when the sequences include not only past but also future data
samples. Interpretation of the trained networks concludes that the
prediction of output features is primarily based on their counterpart in
the input features with the largest emphasis in the sequences being put
on the current time step.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next dates of our Communication Technology Colloquium.
Wednesday, January 6, 2021
Speaker: Raphael Brandis
Time: 10:00 a.m.
Location: https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
Master Lecture: Personalized Occlusion Cancellation in Headphones and Hearing Aids
To users of headphones, hearables, and hearing aids, the occlusion effect (OE) is still a largely unsolved problem that causes them to perceive the sound of their own voice differently. In surveys, users commonly describe this OE-affected own-voice sound as hollow, boomy, and unpleasant. Other body-conducted sounds are affected as well; examples include the noise originating from footsteps or chewing.
A system that actively combats this effect has been in active development at the Institute of Communication Systems. This so-called Active Occlusion Cancellation (AOC) system is designed to cancel the average OE when applied to the average user’s occluded ear. Acoustic measurements as well as a listening test have shown it to be able to significantly reduce the OE and improve the users’ subjective experience. Unfortunately, both the OE and the acoustic characteristics of the occluded ear can vary significantly from user to user. That is why this thesis sets out to explore approaches to personalize the AOC system for individual users.
First, an objective scalar metric for the OE is proposed. This metric is then used to automatically adjust two scalar gains already present in the AOC system. These gains allow for a limited degree of personalization and, up until now, had to be adjusted manually. Second, a technique to iteratively design a personalized feedback controller based on a user’s acoustic measurements is introduced. This feedback controller is a major part of the AOC system. For the limited amount of user measurements that are available, this new design technique manages to quickly and reliably converge onto a controller that meets both the performance and robustness requirements. The controller design process is fully automated and does not require manual intervention by a control engineer. Finally, both personalization approaches are evaluated in a listening test of limited scope and size due to the ongoing COVID-19 pandemic.
and
Wednesday, January 6, 2021
Speaker: Ulrik Deneken
Time: 11:00 a.m.
Location: https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
Bachelor Lecture: Entwurf psychoakustisch optimierter signaladaptiver Übersprechkompensationsfilter
Für die Wiedergabe binauraler Signale über Lautsprecher muss das verwendete binaurale Signal vorgefiltert werden, damit der Höreindruck erhalten bleibt. Das Problem bei der Wiedergabe durch Lautsprechern ist, dass die Kanäle des Signales nicht mehr klar getrennt sind. Signalteile für das eine Ohr erreichen auch das andere Ohr. Dieses Problem wird als Übersprechen bezeichnet. Das Übersprechkompensationsfilter löst dieses Problem, indem die Signale über den Kreuzpfad entfernt werden und die Signale über den Direktpfad möglichst verzerrungsfrei wiedergegeben werden.
Mit Hilfe von psychoakustischer Maskierung soll in dieser Arbeit ein neuer Ansatz für die Entwicklung von Übersprechkompensationsfiltern gegeben werden. Das Signal des Kreuzpfades soll in diesem Ansatz nicht mehr vollständig verschwinden. Es würde ausreichen, wenn das Signal das Kreuzpfades unterhalb der Maskierungsschwelle des Signals des Direktpfades wäre. Hierzu wurden zwei Verfahren ausgearbeitet. Ein Verfahren beruft au fdem Ansatz einer konvexen Optimierung und wird im Frequenzbereich durchgeführt. Das andere Verfahren wird mittels nichtlinearer Nebenbedingungen im Zeitbereich gelöst.
Es konnte gezeigt werden, dass die Filter hinsichtlich der Maskierung verbessert werden können. Dies ist aber mit einem großen Rechenaufwand verbunden und hängt stark von dem verwendeten Eingangssignal ab. Die nichtsignaladaptiven Methoden liefern ein passables Ergebnis, jedoch konnte gezeigt werden, dass eine Verbesserung möglich ist.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Tuesday, December 8, 2020*
*Speaker*: Hendrik Stelzer
*Time*: 10:00 a.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Bachelor Lecture*: Real-Time Use-Case Classification for Intelligent
Control of Hearables
A current goal in the development of headphones and hearables is that
users can wear theam at any time. Especially with the integration of
more and more features, specifically active noise cancellation (ANC) and
transparency features, a method of recognizing different acoustic
situations is necessary to enable intelligent control.
In this thesis an approach for intelligent classification of the current
wearing situation, or use case of headphones and hearing aids is
investigated. In addition to the simple recognition of whether the
earpiece is currently in use, the recognition of special cases such as
different fittings and closing the speaker outlet is examined. The
sensor signals used comprise one external and one internal microphone
signal for each side of the head. Real-time capable features are
extracted from the sensor signals and tested for their information
content regarding a use case classification. Based on headphone
simulations, a forward feature selection is conducted to find the set of
features enabling the most robust classification regardless of any kinds
of disturbane noise. The influence of both air conducted sound (ambient
noise) and body conducted sound (e.g. talking or chewing) is considered.
The mutual information is utilized to measure the quality of a feature.
Five well-known real-time capable classification methods are then
applied using the determined features and compared: the Fisher
discriminant, the least squares discriminant, logistic regression,
decision trees and Gaussian mixture models. Finally, the applicability
of the developed principles is verified by an implementation on a
real-time system (dSPACE ds1005)
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium/
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Thursday, December 3, 2020*
*Speaker*: Timothé Scheich
*Time*: 11:00 a.m.
*L**ocation*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Bachelor Lecture*: Methods for the Assessment of Headphones with Active
Noise Cancellation
Today, Active Noise Cancelling (ANC) headphones are a product daily used
by most people. How goog the headphones sound and how well their ANC is
working with respect to human perception, is still an open research
topic. Only few researchers have been studying the matter and provided
working measurement models. In this thesis, we first lay the groundwork
for a comprehensive understanding of already existing ANC headphones
assessment methods through an extensive research of related work.
Therewith, this thesis also motivates the importance of considering
psychoacoustic as an assessment tool by presenting several metrics
describing pleasantnes and annoyance of a sound. Furthermore, two
measurement series are conducted to gather ANC headphone performance
measurement data using several background noises. The measurement set-up
is explained, and the applied noise databases are presented. Later on,
we propose an assessment model using some of the described metrics.
Finally, we are able to highlight, by comparing the same type of on-ear
headphones, the importance of psychoacoustical parameters that may allow
differentiation of similarly performing objects. We conclude by
explaining the results and propose a subjective test to corroborate our
findings.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear Subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Tuesday, November 24, 2020*
*Speaker*: Fabian Malig
*Time*: 10:00 a.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Bachelor Lecture*: Robust Design of Adaptive ANC Systems Considering
Self-Induced Sound
In addition to passive attenuation, Active Noise Control (ANC) offers a
practical way to reduce the power of disturbing noise. Especially in
modern headphones this technology finds increasing popularity. Adaptive
filters promise a better ANC than time-variant filters since they adjust
to a changing environment. Though, they suffer from high sensitivity
towards noises caused by the user himself, like speaking or impact
sounds. Those noises are called self-induced sounds (SIS) in the following.
This thesis covers the Kalman-Filter as adaptive filter in a feedforward
ANC-system. It outperforms other adaptive algorithms concerning
different disturbances on the system, like self-induced sounds. The goal
of this thesis is the development of noise estimators and methods in
order to reduce the disturbing influence of self-induced sounds on
adaption.
All interested parties are cordially invited, registration is not required.
Generel information on the colloquium, as well as a current list of
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/atktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Friday, November 20, 2020*
*Speaker*: Christian Schlaiß
*Time*: 2:00 p.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master Lecture*: Cancelling of Non-Stationary Disturbances in Active
Headphones
The occlusion effect is a consequence of sealing the ear canals with an
object such as a hearing aid or more generally a hearable. It results in
an amplification of lower frequencies and attenuation of higher
frequencies, which further leads to an unwanted distrotion of the
own-voice perception. To compensate this phenomenon, active and passive
solutions have been presented in literature. Passive solutions include
venting, where a ventilation hole is integrated into the hearing aid.
However, this leads to unwanted feedback. An active approch has been
presented by Liebich et al., in which a time-variant robust feedback
controller compensates the low frequency amplification. This solution is
limited to the attenuation of the occlusion effect on the own-voice.
Other body-conducted sounds, such as footsteps, chewing and swallowing
are not explicitly considered by Liebich et al.
In this thesis, body-conducted sounds exceeding the own voice components
are investigated and tackled. Measurements were conducted to identify
the spectral distribution of different BC sounds due to the occlusion
effect. Afterwards, a controller was designed to attenuate disturbances
in the region of 40 Hz to 80 Hz. This is followed by the introduction of
a stable controller interpolation scheme for switching between
controllers. By use of this so called Youla-Kucera interpolation method,
not only stability but also performance is guaranteed during switching
for a nominal path. Additionally, robust stability for this
interpolation scheme can be proven for discrete steps of delta and
additional constraints on the choice of Q. This is supplemented by a
simple and low complexity approach for detecting footsteps, based on
low-pass filter and recursive smoothing over time. Lastly, the
controller was implemented in real-time on a dSPACE ultra low latency
processing system to validate performance on the footsteps controller
and the switching performance. While the controller worked in theory,
real-time measurements revealed that additional tuning and possibly
additional sensors are needed for a better footsteps controller.
However, the switching scheme showed promising results, fading from one
controller to the other in a stable and smooth fashion. It also revealed
the possibility of real-time controller tuning with the help of the
interpolation coefficient. As all combinations of the two controllers
(with the interpolation coefficient 2) remain robust stable, switching
to different configurations is made possible.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next dates of our Communication
Technology Colloquium.
*Wednesday, November 18, 2020*
*Speaker:* Leonie Geyer
*Time: *2:00 p.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master Lecture*: Sound Field Conversion Using Machine Learning Methods
The reproduction of realisitc sound fields is necessary for the
efficient evaluation of modern communication devices. Depending on the
application, different microphone arrangements are used to record the
sound fields. Not all desired signals are directly available in the
required microphone configuration. The goal of sound field conversion is
to convert any signal between different recording systems in an
identical sound field.
This thesis compares different approaches to convert sound fields. The
conversion between the signals of two concrete measurement systems, a
binaural artificial head and a microphone array with eight channel, is
investigated. Three new approaches, which use artificial neural
networks, are developed. First a convolutional approach, which has a
simple end-to-end structure. Secondly, a time filter approach. Here the
network outputs a FIR filter, which is applied to the input signal.
Third a variant of wavenet, which is divided into to subnetworks. One
analyses a section of the time signal and output a feature vector, which
is available to the conversion network when converting the time signal.
The neural networks are trained using data from real and defined
acoustic environments. The performance is measured by metrics in time
and frequency domain. As a comparison, sound field conversion by
equalising and measuring with the target recording device is performed.
Different experiments are conducted on the structure and
parametrisation, as well as the training process of the neural networks.
Their performance is observed and optimised. The Wavenet variant
achieves in all experiments the best results than the neural approaches.
Training with a loss function, which includes the mean square error and
frequency metrics, can reduce the error in the frequency domain,
although a higher error in the time domain is observed, compared to the
pure mean square error as loss function.
and
*Wednesday, November 18, 2020**
**Speaker:* Shahd Al Hares
*Time*: 3:00 p.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master Lecture*: Signal-Adaptive Approaches to Sound Field Translation
With the growing demands for spatial audio, Ambisonics became to attract
more attention in both fields of recording and reproduction of sound
field. Thus, the demand increased for applying sound field translation
that allows the use to move freely in different directions in the
acoustic scene. In Higher Order Ambisonics (HOA), the sound field
incidence is described in the reference point by a set of mathematical
functions known as Spherical Harmonics (SH). However, the reproduction
is restricted to the surrounding area of the reference point due to the
underlying recording hardware, which can be observed in the Ambisonics
domain as bandwidth limitation. In this thesis, an Ambisonics sound
field translation is investigated. Recent approaches were proposed that
allow the listener to move a few centimeters away from the reference
point (3DoF+). Another approach provides further translation of the
sound field (6DoF) but requires multiple Shperical Microphone Arrays
(SMA) to be used during the recording process.
In this thesis, an enhanced method for sound field translation is
proposed. It is based on upscaling the Higher Order Ambisonics (HOA)
signal to a higher SH order using Compressed Sensing (CS). CS is a
framework that is used to recover a signal from an under-determined
linear system. Different aspects of HOA upscaling and sound field
translation are studied theoretically and practically. Noise reduction
with CS for HOA signals is discussed, and the influence of source
distance on the translated signal is investigated in the
Near-Field-Compensated Higher Order Ambisonics (NFC-HOA) domain. A
systematic comparison between multiple translation approaches, namely
plane wave translation and space warping, is performed based on two
Monte Carlo experiments. Moreover, a new formulae is derived that
defines the limits of the upscaling SH order as a function of the
normalized translation distance and initial SH order. Finally, a method
is introduced for sound field translation of realistic signals based on
upscaling in frequency domain. The method is evaluated in frequency
domain for multiple Bark bands.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Monday, November 16, 2020**
**Speaker:* Lorenz Schmidt
*Time*: 11:00 a.m.
*Locatio**n*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Master Lecture*: Noise Reduction Combining Conventional Approaches and
Artificial Neural Networks
The suppression of noise for single channel speech enhancement is one of
the most prominent challenges in signal processing and has been
addressed for decades. In recent years, the popularization of Machine
Learning algorithms and advances in deep neural network (DNN)
architectures have opened new perspectives and approaches to this field,
yielding impressive results. Many of these algorithms, however, require
a computational effort that exceeds the available resources of a
real-time application. One approach, called RNNoise, combines the
methods of classical signal processing with DNNs. Within a common noise
reduction architecture, a small weighting mask is sufficient to achieve
impressive results. The mask is estimated by a very small neural network
with a low computational complexity.
In this thesis, RNNoise is subject to several modifications that are
intended to improve its denoising performance, while maintaining its
affordable complexity. In a first step, the gated recurrent units (GRUs)
of the RNNoise architecture are replaced simple recurrent units (SRUs),
which improve its performance while speeding up the training process.
The DNN is expanded to estimate the pitch frequency, which is used in
the reconstruction of the harmonics with a comb filter. A new binary IIR
comb filter is developed and added to the signal processing of RNNoise.
Besides the modifications of RNNoise itself, a pitch estimator, based on
ordinary regression, and a mutual information metric are developed. The
evaluation shows a good performance for pitch estimation and voice
activity detection (VAD). A preliminary study analyzes the upper limits,
which can be achieved by the reduced spectral weighting mask. With bark
scaling, 22 gains are a reasonable tradeoff between performance and
complexity. Then, a theoretical evaluation shows that the new network
architecture improves the estimation considerably, especially in
non-stationary noise situations. A final evaluation compares a classical
noise suppression method, an end-to-end neural network approach,
classical RNNoise and the improved model by means of their cepstral
distances, speech-to-noise enhancement and perceptual measures. The
results show that the modifications give the new architecture an edge
over classical RNNoise. On the other hand, the developed IIR binary comb
filter falls back in the expectation and does not improve noise
suppression performance.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/
Dear subscribers of the colloquium newsletter,
we are happy to inform you about the next date of our Communication
Technology Colloquium.
*Tuesday, November 10, 2020*
*S**peaker: *Tom Deckenbrunnen
*Time:* 10:00 a.m.
*Location*:
https://rwth.zoom.us/j/97904157921?pwd=SWpsbDl0MWhrWjY1ZkZaeFRoYmErZz09
Meeting-ID: 979 0415 7921
Passwort: 481650
*Bachelor Lecture*: Significance-Aware Filtering for Nonlinear Acoustic
Echo Cancellation
The acoustic echoes arising in hands-free applications for mobile speech
communication contain considerable nonlinear distortions of the far-end
signal. Some causes of these nonlinearities include the progressive
miniaturization of components and the need for high play-back volume,
two opposing paradigm. It is therefore necessary to use methods of
nonlinear acoustic echo cancellation to provide sufficient quality of
communication. However, the limited resources in mobile devices render
many of these methods to alleviating the high computational complexity.
One approach to alleviating the high computational demand is the
so-called Significance-Aware filtering. Specifically, the cascaded
models based on parallel Significance-Aware decomposition of the system
identification task are treated in this thesis. The aim of this thesis
is the evaluation and enhancement of those structures. In particular,
methods for incorporating nonlinear memory into the parallel
Significance-Aware models are proposed. Beyond that, an analysis of the
necessary adaptation control for these structures is offered.
Results show that better performance can be achieved by the addition of
nonlinear memory to the parallel Significance-Aware models. It is
furthermore shown that the proposed enhancements solve a specific
problem of adaptation control. Nevertheless, adaptation control remains
a major point of interest for the parallel Significance-Aware models.
All interested parties are cordially invited, registration is not required.
General information on the colloquium, as well as a current list of the
dates of the Communication Technology Colloquium can be found at:
http://www.iks.rwth-aachen.de/aktuelles/kolloquium
--
Irina Ronkartz
Institute of Communication Systems (IKS)
RWTH Aachen University
Muffeter Weg 3a, 52074 Aachen, Germany
+49 241 80 26958 (phone)
ronkartz(a)iks.rwth-aachen.de
http://www.iks.rwth-aachen.de/