Digital Signal Processing and System Theory

Talk Torsten Dau


thorsten_dau
Human Auditory Signal Processing in Complex Acoustic Environments

Date: 21.01.2013, 17:15 h - 18:00 h
Room: Aquarium

Prof. Dr. Torsten Dau
Technical University of Denmark, Department of Electrical Engineering
Centre for Applied Hearing Research, Copenhagen, Denmark


Details



In everyday life, the speech we listen to is often mixed with many other sound sources as well as reverberation. In such situations, people with normal hearing are able to almost effortlessly segregate a single voice out of the background – a skill commonly known as the 'cocktail party effect'. In contrast, hearing-impaired people have great difficulty understanding speech when more than one person is talking, even when reduced audibility has been fully compensated for by a hearing aid. As with the hearing impaired, the performance of automatic speech recognition systems deteriorates dramatically with additional sound sources. The reasons for these difficulties are not well understood. This presentation highlights recent concepts of the monaural and binaural signal processing strategies employed by the normal as well as impaired auditory system. The aim is to develop a computational auditory signal-processing model, capable of describing the transformation from the acoustical input signal into its "internal" (neural) representations. Several stages of processing, including cochlear, midbrain and central stages, are considered to be important for a robust signal representation, and a deficiency in any of these processing stages is likely to result in a deterioration of the entire system’s performance. A state-of-the-art model of auditory signal processing would be of major practical significance for technical applications, in digital hearing aids, cochlear implants, speech and audio coding, and automatic speech recognition.

Talk Oliver Mittag Intermediate


olmi    Digitale Simulation von Gitarrenverstärkern

Date: 15.03.2013, 13:00 h - 14:00 h
Room: Aquarium

Oliver Mittag
CAU, Kiel, Germany


Details


Ein Problem bei der digitalen Simulation von Gitarrenverstärkern ist der komplizierte innere Aufbau des Verstärkers, der es schwierig und zeitaufwendig macht, ein Modell aus den tatsächlichen physikalischen Prozessen zu entwickeln. Es ist deshalb zweckmäßig ein Black-Box Modell zu verwenden, bei dem über Eingangs- und Ausgangsmessungen die Modellparameter geschätzt wird. Besondere Schwierigkeiten bereitet es hierbei, die durch Verstärker erzeugten nichtlinearen Verzerrungen nachzubilden. 

Diese Arbeit konzentriert sich auf die Simulation der Nichtlinearität und der dahinter geschalteten Lautsprecherbox. Die Lautsprecherbox wird als Nachfilter modelliert und über gemessene Impulsantworten identifziert. Die Parameter der Nichtlinearität werden auf Basis eines Volterrareihen-Modells geschätzt. Es folgen mögliche, weitere Untersuchungen der Nichtlinearität mit vereinfachten Modellen, die Implementierung in den bereits bestehenden Algorithmus zur Nachbildung nichtlinearer Verstärker-Lautsprecher-Systeme in KiRAT und die Evaluierung. Ein vereinfachtes Modell ist beispielsweise das unten dargestellte Hammersteinmodell, welches bereites echtzeitfähig in KiRAT implementeiert ist.
 
talk_2013_olmi_intermediate_hammerstein.jpg


Talk Frederik Duchâteau


frdu    Parametrierung der geräuschabhängigen Verstärkungssteuerung eines Fahrzeuginnenraumkommunikationssystems

Date: 22.03.2013, 13:00 h - 14:00 h
Room: Aquarium

Fredrik Duchâteau
CAU, Kiel, Germany


Details



Diese Arbeit beschäftigt sich mit der Erfassung, Auswertung und Bewertung der situationsabhängigen akustischen Umgebungsbedingungen, die die wesentlichen Einflussgrößen auf ein Fahrzeuginnenraumkommunikatonssystem darstellen und entwickelt Verfahren zur Ableitung der Parametrierung für die geräuschabhängige Verstärkungssteuerung dieses Systems. Die untersuchten wesentlichen Einflussgrößen sind dabei
5px_blank
  • Geschwindigkeitsabhängigkeit der Hintergrundgeräusche im Fahrzeuginnenraum
  • Besetzungsabhängigkeit des Rückkopplungsverhaltens der Lautsprecher und Mikrofone
  • Temperaturabhängigkeit der elektronischen Bauteile
  • Dämpfung des Direktschalls
Zur Erfassung der Messdaten beliebiger Fahrzeuge wird ein mobiles Aufnahmesystem entwickelt, realisiert und eingesetzt.

talk_2013_frdu_noise_vs_speed.jpg

Die Abhängigkeit des Hintergrundgeräusches von der Fahrgeschwindigkeit wird anhand der aufgenommenen binauralen Audiodaten von zwei Sitzplätzen eines Fahrzeugs bestimmt. Als Beispiel sind in obiger Abbildung Geschwindigkeit und Geräuschpegel einer Testfahrt dargestellt. Analysen wurden gehörrichtig und für verschiedene Fahrzeugtypen durchgeführt.


Talk Oliver Mittag

 
olmi    Digitale Simulation von Gitarrenverstärkern

Date: 24.04.2013, 13:00 h - 14:00 h
Room: Chair library

Oliver Mittag
CAU, Kiel, Germany
 
 

Details

 
 
Heutzutage sind hochwertige Gitarrenverstärker immer noch mit Elektronenröhren aufgebaut, da sich der gewünschte Klang nur schwer mit der modernen Transistortechnik erzielen lässt. Röhrenverstärker haben aber auch ihre prinzipbedingten Nachteile, wie Größe, Gewicht und das Aufwärmen vor der Benutzung, damit sich der gewünschte Klang einstellt. Deshalb gibt es immer mehr digitale Simulationen, seitdem auch die nötigen Signalprozessoren günstig verfügbar sind. Die Probleme bei der digitalen Simulation von Gitarrenverstärken sind:
 
  • Komplizierter innerer Aufbau des Verstärkers,
  • Nichtlineares Ein-/Ausgangsverhalten.

Diese Probleme machen es sehr schwierig und zeitaufwändig aus den tatsächlichen physikalischen Prozessen im Inneren des Verstärkers ein Modell zu entwickeln, mit dem dieser digital simuliert werden könnte. Es ist deshalb zweckmäßig auf ein Black-Box Modell zurückzugreifen. Hierbei ist kaum Vorwissen über das zu identifizierende System von Nöten und die Systemparameter werden mit Hilfe von Eingangs-/Ausgangsmessungen geschätzt.


 
Im Rahmen dieser Bachelorarbeit wurden unter anderem Volterramodelle für verschiedene Verstäkrereinstellungen geschätzt. Obige Abbildung zeigt ein gemessenes Ausgangssignal eines Gitarrenverstärkers bei sinusförmiger Anregung (blau) sowie das durch ein Volterramodell simulierte Ausgangssignal (grün).
 
 
 

Talk Muthuraman Muthuraman


mm
Oscillations in the Brain

Date: 03.06.2013, 17:15 h - 18:15 h
Room: Aquarium

Dr. -Ing. Muthuraman Muthuraman
CAU, Kiel, Germany,
Department of Neurology



Details



In the brain when a large number of neurons are active and firing synchronously it induces large scale macroscopic oscillations. These oscillations can be recorded on the scalp using electroencephalography (EEG) and magnetoencephalography (MEG). The oscillatory activity are observed at different levels (i.e. cellular or neuronal) using different brain imaging techniques which plays a vital role in understanding the information processing in the brain. The rhythmic activity frequency bands are subdivided into delta (1-4 Hz), theta (4-7 Hz), alpha (7-13 Hz), beta (13-30 Hz) and gamma (30-100+ Hz). All these oscillations are generated during different processes and at different regions in the brain. Getting insights on these oscillations can help us in preventing and treating certain brain diseases. In this talk, the analysis of these frequency oscillations is presented. The method used to characterize them is the power spectrum which gives information about the frequency and amplitude. The coupling between two signals can be extracted using coherence, and the dynamics of the power and coherence is obtained using time frequency analysis. A beamforming approach which applies a spatial filter is used to identify the regions in the brain responsible for these oscillations. The network of interaction between these source regions give us more insights about their information flows between the regions which are responsible for certain processes in the brain.



Talk Janika Puls


jap
Automatische Verifizierung von kognitiven Beeintraechtigungen waehrend epileptischer Aktivitaet

Date: 20.06.2013, 13:30 h - 14:15 h
Room: Aquarium

Janika Puls
CAU, Kiel, Germany,




Details



Clinical monitoring is important to adjust the medication of patients, who suffer from absence epilepsy. Monitoring lasts a few hours and is done by the clinical personal, who observes the EEG measurement and asks the patient questions if an absence is detected. This method creates several problems. The EEG data is not observed during the questioning part, and question and answer are not recorded. This thesis presents a program/method for the automatic detection of absence epilepsy and verification of the cognitive impairment during an absence seizure in real-time.

Therefor the signals of the Electroencephalogram (EEG) are analyzed by extracting different features, which were developed according to the characteristics of the absence seizures. The first feature is the short-term power, which results in high values during an absence and low values during normal EEG-activity. Permutation entropy is another feature, which tracks the dynamical changes of EEG data and presents similar results as the kurtosis. The feature kurtosis is based on the fourth and second moment of a random variable and compares the distribution of a random variable with the normal distribution.

The combination of these features allows a robust detection of absence seizures by using a classification method. In this theses two different classification methods are presented: a threshold method and a codebook. These two methods are evaluated with real EEG data of absence patients by computing the classification errors for absence and non absence data and total classfication error. Both of these methods result in a low classification error.

Once an absence is detected the verification of the cognitive impairment is done by asking the patient a question, which is played by a loudspeaker. Via a microphone the answer of the patient as well as the EEG signals are recorded. The recording is done a few seconds before the absence starts until a few seconds after the absence has ended. This allows a further evaluation of the absence seizures.



Talk Grant Davtjan


grant_davtian
Identifizierung und Unterdrueckung von Artefakten in EEG-Signalen in Echtzeit

Date: 20.06.2013, 14:15 h - 15:00 h
Room: Aquarium

Grant Davtjan
CAU, Kiel, Germany,



Details



It is well known that Electroencephalographic (EEG) recordings are widely used in neuroscience in the diagnosis and source localization of many neural disorders and phenomena, such as coma, epilepsy, and tremors. The recorded EEG signals are very often  contaminated with artifacts of physiological (e.g. eye blinking, eye movement, heart beating, and muscle artifacts) or technical (e.g. power-line artifact, electrode popping) origins. Usually,  such artifacts are detected by visual inspection. However, experience is needed in order to know how the artifacts look-like, and hence to mark where and when they appear. Moreover, this task might be timeconsuming, if many channels and many seconds have been recorded. In the case of EEG recordings from epilepsy patients, muscle artifacts are commonly present and contaminate the signals obscuring the desired information; therefore, an efficient filtering technique is necessary. The purpose of the work described in this thesis is to detect and suppress muscle artifacts from EEG data in an automatic manner and in real-time. To achieve that, basic pattern recognition algorithms have been implemented to detect the muscle artifacts. Once they are detected, the signals are enhanced using a Wiener-filter.

Talk Julian Andrej


juan_small

Entwicklung und Analyse eines Echtzeit-Trackingsystems

Date: 04.07.2013, 13:30 h - 14:30 h
Room: Aquarium

Julian Andrej
CAU, Kiel, Germany,




Details



Diver detection is an important factor in harbor protection. Due to the small "sonar cross section" of divers, strong reverberation and many other factors many issues occur. As a result false contacts are generated. In order to maintain robust tracking results automatic tracking and data association is necessary. In this Bachelor thesis a real-time tracking system was implemented an analyzed in terms of computation time and accuracy. The trackings systems was implemented by means of Kalman filter and Sequential Track Extraction within KiRAT (Kiel Realtime Audio Toolkit).



Talk Gerold Baier


Gerold_Baier    The Dynamics of Epilepsy
Date: 02.12.2013, 17:15 h - 18:15 h
Room: Aquarium

Dr. Gerold Baier
University of London, UK


Details



The epilepsies are a disorder of the human brain which manifest as abnormal spatio-temporal rhythms. Recent advances in brain imaging and microsensor  techniques have resulted in high-resolution multiscale measurements in patients, i.e. in vivo. However, the analysis of this complex signals is hampered by the lack of a spatio-temporal model of the underlying brain processes.

Absence

I present recent approaches to the computational modelling of the epileptic rhythms based on nonlinear dynamics of hierarchical brain networks. Unlike animal experiments the computational models can be optimised to the clinical data and predictions can be made that are tailored to the individual patient. I will also argue that the representation of epileptic brain signals as spatial audio allows an intuitive access to these intricate rhythms.




Talk Lukas Goßmann


lukas_gossmann

Development of an Autonomous Lombard Speaker

Date: 15.08.2013, 13:30 h - 14:30 h
Room: Aquarium

Lukas Goßmann
CAU, Kiel, Germany,




Details



The characteristics of human speech signals (e.g., speech l evel or pitch frequency) are highly dependent on the environment of the speaker. When a speaking person is located in an environment with a high background noise, she/ he will start to change these characteristics in order to reach better intelligibi lity. The task of speech processing systems comprises, for example, noise suppression or the improvement of speech intelligibility in general. For this reason, these systems often operate in noisy environments, in which also the mentioned Lombard effect is developed by hu- man beings. In order to optimize or evaluate such speech processing systems, it is reasonable to excite the system always with reproducible noise and speech signals. As a reproducible noise scenario is already available in our audio lab, the aim of this thesis is to create an au- tonomous Lombard speaker using an artificial head with integrated ear microphones and mouth loudspeaker. In a first step, the present noise scenario should be estimated. Based on this estimati- on, the corresponding speech signal is selected and reprodu ced in an appropriate way. For this purpose it is necessary to collect speech and noise d ata for comparison and to develop an algorithm within the Kiel Real-Time Audio Tool kit (KiRAT) using the programming language C. Further enhancement of the Lombard speaker could be to use artificial speech which is adapted with regard to the Lombard effect. For this purpose it is necessary to analyze the characteristics of the Lombard e ffect and apply these in an appropriate way to the artificial speech sign.



Talk Merikan Koyun


meko

Complexity Management in a Real-Time, Ultra-Wideband, Multi-Channel Audio Conferencing System

Date: 01.08.2013, 13:30 h - 14:30 h
Room: Aquarium

Merikan Koyun
CAU, Kiel, Germany,




Details



Conferencing systems are becoming increasingly popular and are frequently used not only in professional work environments, but also in households. Not just reliability and low latency are key factors for these systems, but also the audio and video signal quality and especially the computational complexity for the devices to run the accompanying software. This thesis introduces approaches to lower the computational complexity and latency in a realtime, ultra-wideband, multi-channel audio transmission, without losing audible quality of the signal. It deals with the sound processing of the ultra-wideband, multi-channel, conferencing system by means of filter bank subband reduction and reconstruction techniques. The implementation of a first approach on complexity reduction in terms of a selective subband reduction and reconstruction by linear interpolation in the existing conferencing system is presented along with some results and evaluation. The presentation will include a live demonstaration of the system.



Talk Hamza Bakkari

 
Zwei Studenten und eine Tasse Kaffee   Automotive Patched Array Sensor Model for Self-Calibration and a Precise DoA Estimation

Date: 19.09.2013, 13:30 h - 14:30 h
Room: Aquarium

Hamza Bakkari
CAU, Kiel, Germany,

 
 
 

Details

 
 

In the automotive safety context, radar sensors evaluate the environment to assist the driver. This framework compels a need of precise and consistent estimates of the relative angular positions of present targets in the sensor Field of View. To enable accurate estimation, sensor calibration is compulsory.

Calibrating a radar sensor is or is based over the estimation of its array manifold. The array manifold is the collection of ar- ray responses to signals impinging from all possible Directions of Arrival (DoA).

The purpose of this study is to model the given automotive patched array to enable a blind (e.g. without calibration data) online manifold estimation after a rough initializa- tion (using calibration data) in the scope of a high resolution DoA estimation.

The proposed array model is based on the projection of the complex phase error over the Fourier basis appearing as a nat- ural extension of an antenna spacing mis- estimation concept. After a rough initialization, the defined array parameters are further estimated jointly to the DoA –through conventional beamforming – considering different fil- ters : the recursive least square, the sequential Kalman filter and its single precision adapted version according to Potter´s equations.

 

Talk Simon Ohlendorf

 
Zwei Studenten und eine Tasse Kaffee   Entwicklung einer skalierbaren Sprachaktivitätsdetektion

Date: 19.09.2013, 17:00 h - 18:00 h
Room: Aquarium

Simon Ohlendorf
CAU, Kiel, Germany,

 
 
 

Details

 
 

Mit dem Aufkommen der digitalen Signalverarbeitung wurden in den vergangenen Jahrzehnten zahlreiche neue Anwendungen zur Sprachübertragung und Sprachverarbeitung entwickelt. Mit den schneller, kleiner und vor allem günstiger werdenden Prozessoren und den leistungsstärkeren Übertragungskanälen wie Breitbandinternet und GSM hielt die digitale Sprachverarbeitung auch Einzug in etliche Anwendungen im privaten Bereich, wie z.B. die mobile Telekommunikation, Freisprecheinrichtungen, automatische Spracherkennung, oder Videokonferenzsysteme.

Da Sprachsignale auch Pausen enthalten, lässt sich bei der Verarbeitung von Sprache Rechenaufwand und Übertragungskapazität einsparen, wenn diese Pausen erkannt werden. Dies geschieht anhand einer sogenannten Sprachaktivitätsdetektion (Voice Activity Detection, VAD). Je nach Umgebung ergeben sich verschiedene Problemstellungen an die VAD.

In dieser Arbeit soll der Einsatz einer VAD für ein Videokonferenzsystem in einer Büroumgebung untersucht werden. Es soll ein Verfahren entwickelt werden, welches die Sprachanteile eines Signals in der Büroumgebung zuverlässig detektiert. Hintergrundgeräusche wie Mausklicks, Tastaturgeräusche, Lüfterrauschen der Computer usw. sollen als Hintergrundgeräusch klassifiziert werden. Die VAD wird skalierbar sein, indem verschiedene Methoden für die Detektion angewendet werden, welche sich je nach gewünschtem Rechenaufwand (de-)aktivieren lassen können, je nachdem, ob Wert auf eine möglichst exakte Entscheidung oder auf Recheneffizienz gelegt wird.

 
 

Talk Wiebke Schmidt


wsch
Noise Reduction in EEG and MEG Signals

Date: 25.09.2013, 13:00 h - 14:00 h
Room: Aquarium

Wiebke Schmidt
CAU, Kiel, Germany,




Details



Magnetoencephalography (MEG) and electroencephalography (EEG) denote the measurements of currents within the brain. Currently the most used method for recordings is EEG. Due to the fact that the MEG sensors have to be liquid helium cooled, MEG recordings are very expensive. However, MEG offers some advantages in comparison to EEG. One of the advantages is that the measurement can be done without contacting the scalp. Therefore, the time consumption for the setup of the measurements is considerably smaller than EEG.

With the help of EEG and MEG, many brain disorders and also the area of damage can be identified. That is very important in the Medical Technology area. There are many applications where the brain signals can be used to. For example, electronics such as the television can be controlled only with thoughts.It can also be applicable in cars to avert, for example, an accident,
because the brain knows several milliseconds earlier, that the car should be braked. The sensors could be assembled in the headliner of the car. By a connection of the sensors with the car electronic, the car will brake before the signal with the information to brake arrived the leg.      

There are new magnetoelectric (ME) sensors that are being developed at Kiel University. These sensors are cheaper and they can be operated at room temperature. As a result, MEG examinations can be done with lower costs. However, the main problem is that the measurements present noise from different sources (amplifier, sensor itself, computers etc.) which may hide the brain signals. For that reason, an enhancement has to be done.

The purpose of this Thesis is to reduce the noise from the EEG/MEG signals. To achieve that, the sensor behavior was simulated and an algorithm to estimate the noise and then to enhance the signals were implemented in MATLAB. 



Talk Mirco Scheil


misc
Development of an Audio Upmix Scheme for Spatial Playback of Stereo Sources

Date: 07.11.2013, 13:30 h - 14:00 h
Room: Aquarium

Mirco Scheil
CAU, Kiel, Germany,




Details



While 5.1 surround multichannel audio systems are being adopted widely by consumers, most audio content is still available in the two-channel stereo format. For audio systems enhancing the sound experience beyond stereo, it is thus crucial that stereo audio content can be played back, desirably with an improved experience compared to the legacy systems.

The benefit by using more front loudspeakers will be an increased width of the virtual sound stage and an extended sweet-spot region. Lateral independent sound components can be played back seperately over loudspeakers on the sides of a listener to increase listener envelopment.

sourround_lsp_placement

 In this thesis a real-time upmix-algorithm will be developed within the Kiel Real-Time Audio Tool kit (KiRAT) using the programming language C. The aim will be to play back stereo signals over more than two loudspeakers for an improved listener experience. After using analysis filterbanks a spatial decomposition of the two-channel audio signal will be performed with correlation- and coherence-based analyses. Given the spatial decomposition of the stereo signal, the single loudspeaker signals can be generated.



Talk Stephan Senkbeil


stse

Feature Extraction and Pattern Recognition for Automatic Speech Recognition (in German)

Date: 07.11.2013, 12:30 h - 13:30h
Room: Aquarium

Stephan Senkbeil
CAU, Kiel, Germany,




Details



The aim of this thesis is to design and implement a basic system for speech recognition. In order to do so, a feature extraction has been implemented which calculates Mel Frequency Cepstral Coefficients (MFCCs). These features are made more robust by using a cepstral mean normalization which removes the influence of the used microphone and the acoustic characteristics of the room on the calculation of the MFCCs.

Following this, a phone recognition in realtime using a GMM was developed and implemented in KiRAT. The estimation of the needed parameters was performed in MATLAB by using the EM algorithm. Alternatively, the TIMIT-database or the „Kiel Corpus of Spontaneous Speech“ were used as data source. The evaluation of the recognition was performed by using a confusion matrix. A recognition rate of 39.3215% could be achieved.

An isolated word recognition had been realized as an additional model using multiple HMMs. For the recognition, „left-right“ models with a word-dependent number of states were used. The vocabulary of the recognizer is organized in so-called sets and can be compiled in KiRAT . The training of the models is also done in KiRAT. Just as the phone recognition, the isolated word recognition has been evaluated. For the use of three states per phone of each word and a vocabulary of ten words a recognition rate of 79.3758% could be reached.

As an extension of the single word recognition, a connected word recognition has been implemented in order to control KiRAT by voice. For the purpose of training models for the command words for the voice control, speech recordings were done. An evaluation of the recognition rate of the connected word recognition has not yet been performed but for a limited vocabulary and just four commands a subjectively good performance could be observed.



Talk Michael Brodersen


mibr

Real-Time Signal Enhancement for Breathing Masks

Date: 21.11.2013, 13:00h - 14:00 h
Room: Aquarium

Michael Brodersen
CAU, Kiel, Germany,




Details



If todays firefighters want to communicate with each other while wearing a breathing mask during fire-fighting or rescue operations, there are numerous technical aids available. Modern breathing masks are equipped with microphones and loudspeakers (inside and outside the mask) and also a connection to wireless devices is state of the art. In this way, firefighters can communicate with each other via radio or also directly by recording the speech signal within the mask and playing it back via the loudspeakers located outside the mask. Due to the reproduction of the speech signal via the loudspeakers the signal couples back to the microphone inside the mask. This can cause a disturbing feedback, which might lead to an instable system. The signal quality and also the disturbing feedback can be significantly improved by the use of digital signal processing. Therefore, the aim of this thesis is to create a real-time signal enhancement system for three different communication paths

  • from the radio to the loudspeaker inside the mask (head phone),
  • from the microphone to the base station of the radio and
  • from the microphone to the loudspeakers outside the mask.
Due to these different communications paths, different algorithmic approaches should be applied to overcome problems like background noise, breathing noise, varying speakers, etc. To realize these algorithms an existing real-time framework (KiRAT) can be utilized, which already provides usefully features like for example an audio connection or a filter bank.

 



Talk Celina Schneider


cesc

Evaluation of Speech Enhancement Algorithms

Date: 20.11.2013, 11:30 h - 12:30h
Room: Aquarium

Celina Schneider
CAU, Kiel, Germany,




Details



Heutzutage übernimmt die digitale Übertragung von Sprache immer mehr Relevanz. In vielen Ger¨aten werden Algortihmen implementiert, die die Sprachverständlichkeit verbessern und somit nicht nur einen höheren Komfort bietet, sondern auch mehr Sicherheit, da zum Beispiel im Auto der Fahrer sich bei einer Unterhaltung nicht mehr umdrehen muss. Die Aufgabe von Sprachbearbeitungs-Algortihmen besteht in vielen Anwendungen darin eine Geräuschunterdrückung zu erzielen oder die Verbesserung der Sprachverständlichkeit im Allgemeinen umzusetzen. Der Lehrstuhl für Digitale Signalverarbeitung und Systemtheorie befasst sich unter anderem mit Verbesserung der Kommunikation in Kraftfahrzeugen. Dazu steht eine S-Klasse in der long Version der Marke Mercedes Benz zur Verfügung. In diesem Fahrzeug sind verschiedene Mikrofone und Lautsprecher eingebaut, die für den Anwendungsbereich des Freisprechens und der Fahrzeuginnenraumkommunikation Verwendung finden. Zur Optimierung der Fahrzeuginnenraumkommunikation, auf englisch In-car communication (ICC), hat die Arbeitsgruppe schon einige Erfolge erzielen können. Das ICC System dient zur Verbesserung der Kommunikation in einem fahrenden Auto. Dazu sind die Lautsprecher an den Fahrzeugtüren und in der Hutablage eingearbeitet. Bei den verwendeten Mikrofonen handelt es sich um Gurtmikrofone, die von einer externen Firma namens Paragon eigens hergestellt wurden. Dabei sind die Mikrofone in den Sicherheitsgurt mit eingewebt worden. Weiterhin befasst sich der Lehrstuhl für Digitale Signalverarbeitung und Systemtheorie mit der Verbesserung der Kommunikation von Einsatzkräften der Feuerwehr, während sie eine Atemschutzmaske tragen und in einem Einsatz tätig sind. Die Kommunikation während eines Einsatzes erweist sich als äußerst problematisch. Die Kommunikationseinheit ist in diesem Fall direkt an die Atemschutzmaske gekoppelt. Sie befindet sich an der Sauerstoffzufuhr, sodass nicht nur die Umgebungsgeräusche ein wichtiger Störfaktor sind, sondern auch das Ein- und Ausatmen einen sehr großen Bestandteil der Störgrößen darstellen. Eine Atemschutzmaske ist demnach nicht nur mit der wichtigen Sauerstoffzufuhr und wichtigen Filtern ausgestattet, sondern auch mit zwei Lautsprechern und einem Mikrofon. Die Lautsprecher sind außen an der Atemschutzmaske angebracht, sodass ein Feuerwehrmann/frau ohne einen Funkpfad kommunizieren kann. Dies ist besonders wichtig, wenn mit einem Verletzten bzw. mit einem Opfer gesprochen wird und somit vergleichbar mit einer Freisprecheinrichtung eines Autos ist.

Nach der Umsetzung dieser Systeme stellt sich die Frage, inwiefern sich die Qualtität eines solchen Sprachverbesserungssystems bestimmen lässt. Dies wird das Ziel dieser Bachelorarbeit darstellen. Eine M¨oglichkeit wäre verschiedene Probanden Audiodatein, die einen Sprachverbesserungsalgorithmus durchlaufen haben und welche die keinen durchlaufen haben, vorzuspielen und nach eigenem Empfinden subjektiv bewerten zu lassen. Die zweite Variante wäre, anhand einer objektive Methode die Systeme zu evaluieren. Dazu würden verschiedene Methoden in Matlab implementiert, dann die Distanzen berechnet werden und nach eigener Erwartung bei einem optimalen Algorithmus würde die Distanz möglichst klein bleiben. Eine weitere Möglichkeit einen Sprachverbesserungsalgorithmus zu evaluieren, wäre die Betrachtung des Verhältnisses von Sprache und Rauschen (SNR) der Signale vor und nach dem Sprachverbesserungsalgorithmus. Der Hauptaugenmerk dieser Bachelorarbeit wird dabei auf dem In-car-communication System liegen.

 



Talk Thi Thu Hien Vu


hien_2
System Design and Implementation for Simulating Time Variant Acoustics Channels in Real Time

Date: 05.12.2013, 13:00 h - 14:00 h
Room: Aquarium

Thi Thu Hien Vu
CAU, Kiel, Germany,




Details



A transmission channel is a fundamental element in a communication system, which provides the means to transmit information from a transmitter to a receiver.  In reality, these channels vary with time, which leads to a unstable performance of a communication system. The goal is to test or evaluate new models, algorithms and applications which involve acoustic channels.

Within the scope of this talk, a method to simulate time-variant acoustic channels in real-time is proposed, in which time-variant acoustic channels have been generated in different environments scenarios and with a specific number of moving reflectors. Furthermore, with the efficient convolution and complex impulse response generation method, the computation complexity has been improved in comparison to the formal usages.



Talk Thi Thu Hien Vu


hien_2 System Design and Implementation for Simulating Time Variant Acoustics Channels in Real Time

Date: 05.12.2013, 13:00 h - 14:00 h
Room: Aquarium

Thi Thu Hien Vu
CAU, Kiel, Germany,



Details


Diver detection is an important factor in harbor protection. Due to the small "sonar cross section" of divers, strong reverberation and many other factors many issues occur. As a result false contacts are generated. In order to maintain robust tracking results automatic tracking and data association is necessary. In this Bachelor thesis a real-time tracking system was implemented an analyzed in terms of computation time and accuracy. The trackings systems was implemented by means of Kalman filter and Sequential Track Extraction within KiRAT (Kiel Realtime Audio Toolkit).