GaAs MMIC are taking big space in the wireless communication in today’s world. The devices are to be characterized for higher frequencies. External passive waveprobes are used to characterize the devices. As the use and technology is increasing there will be more development on MMICs. It would be nice to have a passive waveprobe on the chip itself so as to minimize the losses created due to junctions.In this master’s thesis several wave probes designing techniques are described. The wave probes are designed and simulated. Wave probes are loop couplers with a sufficient directivity and very small insertion loss. The proposed wave probes or coupler are realized using TriQuint GaAS pHEMT MMIC process. Simulated data for different wave probes are compared and it is found that a loop coupler at the edge of the transmission line radiating electromagnetic is best suited. The electromagnetic radiation causes coupling or cross talk. Radiation at the discontinuities or edge is high and this advantage is used to design a coupler with a high directivity. In the design process size of the device is tried to be kept as small as possible so as to make the cost low. EM-simulation of passive MMIC have been carried out and compared with the experimental result to validate the use of foundry models with real life measurements. / The author was examined at Norges teknisk-naturvitenskapelige universitet (NTNU), Trondheim

The amount of non-textual media on the Internet is increasing, which creates a greater need of being able to search in this type of media. The goal with this thesis is to be able to do information search by use of soundtracks in audio databases. To get to know the content in an audio file, one wants a system that can automatically extract necessary information. The first step in making this system is to record what is happening at which time in an event log. This thesis treats the beginning of such a process. The experiments performed dealt with detection of pauses lasting longer than 1 second and detection of speaker changes. The corpus used in experiments consists of news broadcasts from The Norwegian Broadcasting Corporation (NRK) radio. Each broadcast had a transcription, which was used as a reference when evaluating the results. Another corpus, the HUB-4 1997 evaluation data, was used for comparative tests.A lot of work treating indexing of audio databases has already been conducted. As corpora are different, there may be varying results obtained from the same methods. In this thesis, common segmentation methods have been used with the parameters adapted to give as good results as possible with the given corpus. In the pause detection, model-based segmentation was used. A Gaussian mixture model was implemented for each of the two events: sound and long pause. For the speaker segmentation, experiments with different metric-based segmentation techniques were performed. The Bayesian information criterion (BIC) and a modified version of this criterion were tested with different options and parameter values. A false alarm compensation based on the symmetric Kullback-Leibler distance was implemented as an attempt to reduce the number of false change points. The pause detection was not successful. By using the manual transcription as reference, an F-score of 38.1 % was obtained when the settings were adjusted to result in about the same numbers for false alarms and false rejections. However, further investigation showed that the transcription had flaws with respect to labeling of pauses. An evaluation of the wrongly inserted pauses showed that most of these segments actually contained silence or noise. However, the number of pauses missed was unknown, and it was not possible to get a reliable F-score. An attempt on labeling all pauses in the HUB-4 1997 data was done. With the modified transcription, an F-score of 81.7 % was obtained. However, it is possible that unlabeled pauses still exist in the transcription, as the labeling was performed by only looking at the audio signal. From classification experiments it became clear that using 1st and 2nd order delta coefficients in the feature vectors gave an improvement over just using static MFCCs. An F-score of 98.8 % was obtained from these experiments, which implies that the models are good when the segment boundaries are known. In order to get trustworthy results from the recognition task, a review of the transcription must be done.When using the modified version of BIC and false alarm compensation for speaker change detection, an F-score of 77.1 % were obtained. The average mismatch between correctly detected change points and reference transcription was 339 milliseconds. As a measure of how good the algorithm is, an F-score of 72.8 % was obtained with the HUB-4 1997 data. Ajmera et al. (2002) obtained an F-score of 67 % with the same data. It became clear that full covariance matrices gave an improvement over diagonal covariance matrices and that static MFCCs as feature vectors gave better results than MFCCs including delta coefficients. Inclusion of pitch as another feature did not contribute to any improvement of the results.

Most automatic speech recognition systems are based on statistical models thatrequire training. While these types of systems have reached recognition ratesthat are sufficient for many purposes, they perform poorly for speaker typesthat are not present in the training material. Children are often absent fromtraining material for speech recognizers, and creating good training materialfor children can be difficult and expensive.To address this issue, this thesis focuses on using adult training material totrain a recognizer for children by adapting the training material duringtraining. Instead of performing speaker-dependent adaptation duringrecognition, where computational power may be scarce, and responsiveness may beessential, adaptation is performed during training towards a class of speakers.Using a combination of vocal tract length normalization (VTLN) and cepstralmean normalization during training, promising results have been obtained. In aconnected-digits task, a reduction in errors as high as 70% was shown, with areduction of almost 50% in a large vocabulary task. Using VTLN to warp thesame training material several times, combining these warped materials to trainone recognizer, a similar reduction in errors was shown, but with an increasedrobustness indicating a less speaker-dependent system. It is also shown that apiecewise linear warping method is better suited to warp adult speech to childspeech, than a bilinear warping method.

This report presents a potential method to efficiently implement edge diffraction from a noise barrier into geometrical room acoustic softwares. The modelling is based on semi-transparent surfaces, and the classic digital signal processing multipath transmission equation has been employed to describe in a mathematical term the presented method. The basic idea is to subdivide the noise barrier into a number of subareas, and then give each subarea an optimalized transmission coefficient for building the best possible output impulse response.To evaluate the proposed semi-transparent modelling, a Matlab simulation model of an infintite noise barrier case has been developed, and the corresponding simulations have been compared with the ideally correct solution. Accordingly, it is stated that there seems to be a clear positive potential in the proposed modelling technique. However, the results also reveal a somewhat instability in the modelling, which is expected to appear mainly for rare critical source and receiver positions.A main goal has been to develop a method that can easily be implemented in the existing calculation algorithms of today's commercial software developers. For verification, the proposed modelling has by discussion been associated with the often employed diffuse rain method. However, since no true implemenation in geometrical room acoustic software has been performed, further studies are required.To maintain efficiency and reliability, another desired outcome of the presented modelling has been that is should function for a general one-to-all source-receiver condition. Surely, the modelling seems to give fairly good results for symmetric source and receiver positions, but as the receiver is moved away from these symmetric conditions some unwanted errors occur, especially at higher frequencies.Main focus has been given to receiver positions located in the shadow zone, but some simulations and discussion has also been given to receiver positions located near the source-receiver sight line - at where direct sound energy contributions are also included and an interference pattern arises. To cope with this interference pattern, a polarity shift is proposed, which gives a clear improvement at low frequencies.One certainly interesting feature of the presented modelling technique is that it involves a broadband-based simulation method, which means that it gives the full frequency response by running only one simulation. Indeed, this is advantageous regarding calculation efficiency, but it does however also introduce some issues regarding a potential future software implementation - as the common case in geometrical room acoustics is to run individual octaveband-based simulations.

High frequency dispersion has a great influence on the perceived performance of a loudspeaker. The directivity of a single transducer primarily depends on driver size, however directivity can be modified using an acoustical waveguide. A method of modelling and designing a wide dispersion waveguide for a loudspeaker soft dome tweeter has been developed.A combination of finite element (FE) modelling and understanding of directivity and waveguides is used in order to prototype loudspeakers virtually. By utilizing computer simulations, the prototyping process is faster and more cost effective, all the while designing better performing loudspeakers.Firstly, a baseline acoustic-structure interaction FE model of a tweeter was built in the Comsol Multiphysics software. The model was verified by measurements, and the directional properties showed satisfactory agreement in the frequency range of interest. The accuracy of the baseline model allowed for credible simulations of waveguides.Secondly, many waveguide geometry types were investigated, and a method for randomizing geometries and automating the design process was developed using the Livelink for Matlab module in Comsol. Subsequently, a best fit waveguide design was selected based on a set of defined design criteria.Thirdly, a prototype was built, the measured performance compared to the simulated model, and discrepancies investigated. The waveguide directivity performs as modelled through most of the working range, although deviations from simulations were larger than expected at frequencies above 12 kHz. The measurements validate the modelling procedure and emphasize the value of the design algorithm, even though the prediction accuracy may be improved. It can be shown that a waveguide of this type can, with only small modifications, be an effective way to increase HF dispersion for a large range of commercially available tweeters.

An adaptable binaural model for surround sound has been developed in this master’s thesis. The adaptability is based on measurements of the listener’s head. This model is based on what was found to be the best suited material combination of successful models in earlier studies. This includes an ellipsoidal model for interaural time difference, an one-pole, one-zero head shadow filter and the use of Blauert’s directional bands for spectral manipulation. The model can play back six channel surround content using the standardized 5.1 surround sound loudspeaker setup. This standardized loudspeaker placement is used when creating virtual sound sources. Arbitrary sound directions are made in the horizontal plane by creating virtual sound sources using vector base amplitude panning between the standardized loudspeaker positions.To test the performance of this model, a listening test was conducted. The hypothesis tested was that the adaptable model would produce equal or lower localization error, compared to the commercial model. 20 test subjects participated. The test featured three different test types; standardized 5.1 loudspeaker setup, a commercial model for surround sound in headphones, and the adaptable model. Localization accuracy for ten selected directions in the right half plane was tested. The results from the adaptable model were compared to the result of the commercial model. The loudspeaker setup acted as a reference.Mean localization error was found to be thrice as high for the adaptable model, compared to the commercial model. Both models had the same standard deviation. 95% of the confidence intervals for these models did not overlap, i.e. there is a significant difference between the two methods. With this one can safely conclude that the commercial model provided a smaller localization error than the adaptable model. Hence the hypothesis has to be disproved.Both the commercial model and the thesis model performed significantly worse than the loudspeaker setup. One difference between commercial model, and the thesis model, was that that the commercial model had added room reflections and reverberation. This can create the sensation that the sound is coming from outside the head, and make it easier to localize. This contradicts with the knowledge that reverberation diffuses the sound field, making the direct sound that provides the directional information become less prominent.

This study has investigated to what extent musicians adjust their source levels to different music rehearsal rooms. In the experiment, six amateur musicians were to perform the same song i four different rehearsal rooms, by first singing, then by playing guitar and last by combining singing with guitar playing. All sound sources were recorded and analyzed. The results shows that the average musician adjusts his source levels to the rehearsal room and that most of the adjustments are made in the guitar playing. Looking at the individual musician there are some that do not show any signs as to being affected by the rooms, and there are some that shows clear signs of being affected by the rehearsal room. The result also shows that the musicians are affected differently by different acoustic parameters, whereas the strength shows the least correlation and reverberation time shows the most correlation to the adjustment made by the average musician.

A video conferencing situation combines the acoustical properties of two rooms. As the talker is located in a one room, the sound reaching the listener in an other room will be colored by the acoustical characteristics of both of them. This thesis aims to survey the current conditions in a selection of video conferencing rooms, by investigating several room acoustic parameters involving background noise, re- verberation time, speech clarity and speech intelligibility. Convolution of recorded impulse responses enables the rooms to be combined, and the combined results to be evaluated. The evaluation of the results allow limit values for acceptable quality for video conferencing to be suggested. Limits for the highest acceptable values for the early decay time and the lowest acceptable values for speech clarity are sug- gested both for the single-room situation, and for the combination of rooms. The suggested values are based on specifications from building standards and relations between measured room acoustic descriptors.

The objective of this assignment was to implement a bistatic radio acoustic sounding system (RASS) in a anechoic chamber, using equipment at hand. The purpose of which was to enable measurements within a controlled environment to avoid variations in wind, humidity and temperature. Radio acoustic sounding systems as well as planning of the test setups have been theoretically investigated in previous work.The approach to the task has been to choose equipment assumed suitable for the experiment and verify the equipment suitability by measuring the technical characteristics relevant to the task. Then a bistatic RASS was constructed within a anechoic chamber. Using software defined radios, the received signals were digitally sampled. So was the acoustic field in the volume of interaction.Although processing the results failed to detect any sign of scatter, the equipment has been thoroughly tested. The acoustic, as well as electromagnetic equipment have shown good results with respect to the suitability for such a system prototype.No conclusion as to whether any scattering occurs as a consequence of the acoustic field can be made, as the absence of such indications may merely indicate insufficient sending power or too high levels of background noise. Prior to these findings, the equipment were concluded to be suitable for a prototype bistatic RASS. Methods to improve the measurement quality have been suggested as future work.

In this report various techniques used to estimate head related impulse responses are compared. The purpose is to investigate the effectiveness of presenting auralization via the QuietPro system's earplugs, and see if sound localization in the horizontal plane is possible. In addition, a theoretical study to relate pinna dimensions to features found in measured head related impulse responses is described. In the theoretical part, impulse responses for 33 left ears found in the CIPIC database were investigated, as an attempt to relate reflection coefficients and time delays associated with reflections from the pinna to physical dimensions of the ear. Unfortunately, no clear connection was found.In the listening test, the participants were sitting in the middle of a circle, surrounded by 36 numbered pieces of paper (either standing on top of e.g loudspeakers or attached to microphone stands) that indicated possible sound directions. 14 subjects performed sound localization tests by listening to three consecutive noise bursts of 150 ms duration with 100 ms silence between. Prior to the experiment, measurements of the subject's head were made and used for customization of the models. The task was to determine which of the 36 possible directions the sound was meant to come from. Seven simulation conditions were evaluated, each including 33 stimuli. Four test stimuli were also presented, resulting in a total of 235 noise bursts for each subject.The results show that the presented methods provide directionality to the stimuli, and that sound localization is possible. However, a significant reduction in localization performance compared to what could be expected for normal hearing conditions is observed. A high number of front/back confusion is reported, and even some instances of left/rigth confusion. Accuracy of the results was not predicted by model complexity, and in some cases it turned out that adding more features significantly degraded the performance.