Reverberation is a key aspect when designing the interior of buildings, and must be carefully considered in the context of the function of the room. Defined by the reverberation time (RT), it is known to have a big influence on the intelligibility and quality of audio in closed spaces.
In this work, we investigate the relationship between the RT and explore the feasibility of using multilayer perceptron (MLP) networks to create an estimator for the RT by using the values of objective measures as input features. We investigate five measures in particular: the Perceptual Evaluation of Speech Quality (PESQ), Virtual Speech Quality Objective Listener (ViSQOL) and its extension focused on audio (ViSQOLAudio), and the Short-time Objective Intelligibility Measure (STOI) and its extension ESTOI.
We create a 3-layer MLP network that estimates the RT with a mean absolute error of 0.144 on our simulated RIR test sets and 0.196 on our real RIR test set.

Estimating reverberation time (RT60) accurately is crucial for enhancing the acoustic quality of various environments as it decides how you feel the sound fades away subjectively. Traditional methods, such as Sabine's equation, require extensive prior knowledge and assume ideal conditions, limiting their practicality. To address these limitations, this paper explores the application of convolutional neural networks (CNNs) enhanced with an encoder architecture based on transformer mechanisms for blind RT60 estimation. The proposed model leverages simulated and real-world datasets, incorporating environmental noise to improve robustness. Results indicate that the CNN-Encoder model achieves superior performance, with a mean squared error (MSE) as low as 0.0006 seconds for pure room impulse responses (RIRs) and 0.0011 seconds under +30dB signal-to-noise ratio (SNR) conditions. It also demonstrates potential in practical usage achieving an MSE of 0.0282 seconds under audio recordings. This approach offers a significant reduction in estimation error compared to the CNN-only architecture, demonstrating the potential for improved acoustic parameter estimation in varied environments. Future work will focus on further optimizing the model for real-world applications and reducing computational complexity while maintaining high accuracy.

In building design, it is important to consider certain materials for certain acoustical properties. Specifically, the time it takes for an audio signal to decrease in volume by 60 dB is important. This can be estimated with Sabine's and Eyring's formula's, which both make use of the average absorption coefficient of the materials in a room. This absorption coefficient indicates how much of the original audio signal is absorbed into the material. However, measuring these absorption coefficients for a material is difficult and time consuming.
In this study, a machine learning approach is used to estimate the absorption coefficients, by using the room impulse response in combination with the layout of a room. A room impulse response is the characterizing sound of a room. These two pieces of data are processed through a convolutional neural network and a multilayer perceptron, respectively, and combined to make the final prediction of absorption coefficients. A novel approach of simulating the data is used, and a real dataset is used in conjunction with the simulations to use a state-of-the-art regression loss function made for domain adaptation. The results show that the machine learning approach still has a large error compared to using Eyring's formula, and that machine learning is not yet a viable option to use instead of conventional methods.

The accurate simulation of Room Impulse Responses (RIRs) is important in a variety of applications in acoustics such as automatic speech recognition, speech enhancement, and architectural acoustic design. While objective metrics for evaluating RIRs have been researched extensively, the subjective perceptual accuracy of the simulations is largely overlooked. This paper seeks to address this gap, designing a subjective testing methodology for evaluating the perceptual accuracy of simulated RIRs. A framework is proposed that combines the ABX testing methodology with a modified Multiple Stimuli with Hidden Reference and Anchor (MUSHRA) approach, measuring attributes such as clarity, warmth, environment, and reverbera- tion. The study involved 50 participants evaluating audio samples convolved with both real and simulated RIRs. Results seem to indicate that participants could reliably distinguish between real and simulated RIRs, with perceptual differences observed in the “clarity” and “reverberation” attributes. The findings suggest that current simulation methods for RIRs do not fully capture the perceptual aspects of acoustic environments.

Automatic Speech Recognition (ASR) systems are becoming increasingly popular in this day and age. Unfortunately, due to inherent biases within these systems, performance disparities exist among specific demographic groups. Bias metrics can be used to measure this bias. Within ASR they represent a niche area that has not yet been thoroughly explored. The few bias metrics that exist in literature mainly centre around the performance differences between speaker groups. This paper proposes two new bias metrics that focus not only on performance differences, but also take the base performance into account: Weighted Performance Bias (WPB) and Intergroup Weighted Performance Bias (IWPB). Although the lack of ground truth makes the results less easily interpretable, the results show similar trends within the new metrics as those defined in literature: bias is greatest among non-native Dutch speech.

Automatic Speech Recognition (ASR) systems have become an integral part of daily lives. Despite their widespread use, these systems can exhibit biases that express themselves in the differences in their accuracy and performance across different demographic groups. Methods quantifying these biases have been developed. This paper investigates the relationship between bias and the acoustic characteristics of speakers. By examining various acoustic embeddings, derived from models like wav2vec 2.0 and XLSR, we aim to identify which embeddings correlate most strongly with bias. The findings offer insights into improving the fairness of ASRs by exploring how acoustic features influence bias in ASR systems. Future research directions include exploring isolated speech properties and extending the study to diverse linguistic contexts to deepen understanding in this area.

Dutch State-of-the-art Automatic Speech Recognition (ASR) systems do not perform equally well for different speaker groups. Existing metrics to quantify this bias rely on demographic metadata, which is often unavailable. Recent advances in the field use machine learning to find groups of similar speakers instead. However, its black-box nature obscures the interpretability of resulting groups. This paper proposes an interpretable approach to bias discovery by clustering speakers based on acoustic and prosodic features. Different feature subsets were compared in their ability to find performance disparities in five ASR systems for two separate speaking styles. Results show that these feature sets can uncover bias approaching known disparities between demographic groups. While the effectiveness per feature set differed between the speaking styles, the most successful ones found significant disparities between clusters with diverse demographic compositions.

This paper presents a novel approach to measuring bias in Automatic Speech Recognition (ASR) systems by proposing a metric that does not use the conventional approach of a reference group. Current methods typically measure bias through comparison with a ’norm’ or minimum error group, potentially introducing additional biases. To address this issue, this study introduces a new metric: a combination of the Groupto-Average Log Ratio and the Sum of Group Error Differences. This metric aims to provide a fair comparison by measuring performance relative to the average of the groups rather than a single reference group. Results indicate that the new metric reveals different aspects of bias not captured by traditional methods. This study contributes to the ongoing research on fairness in speech technology by challenging the existing bias metrics and proposing alternatives that might offer more equitable evaluations. Future research should explore further refinements of these metrics, and apply them across more varied datasets and environments. Ultimately, this research moves towards making ASR technologies more inclusive, ensuring that they serve all user groups equitably.

Feature extraction techniques for content-based image retrieval are explored, focusing on black-and-white images in the context of historical watermarks. Orthogonal moments and texture features are found to be most applicable. Seven methods are evaluated: four different orthogonal moments, Gabor features, and two novel combinations of orthogonal moments with Gabor features. Retrieval effectiveness is judged based on the precision-recall curve and mean average precision, and watermarks are considered when unchanged, rotated, sheared and both rotated and sheared. The results demonstrate that research into improving efficient grayscale image representation does not translate over to improvements with black-and-white images. As it stands, the basic Zernike moments and novel Gabor-Zernike features are most effective.

This paper tackles the problem of evaluating the task of finding similar scanned historical watermarks - small images embedded in historical paper that have been digitized to be processed on a computer - using pre-trained neural networks. This research aims to identify an efficient and accurate alternative to the traditional, time-consuming manual detection methods for finding similar watermarks. The primary issue addressed is the inefficiency of these manual methods. The evaluation focuses on finding similar watermarks for a specific query watermark, assessing the efficacy of neural networks in comparison to a prior art system that employs traditional image processing techniques. This comparison aims to determine how well these neural networks perform in the task of watermark similarity detection. The study involves a dataset of 500 labeled images tested in two distinct contexts: one using unprocessed images and another using images processed to keep only the watermark outline. The results show that pre-trained models achieve higher accuracy and time efficiency compared to the prior art system that uses image processing. These models demonstrate significant effectiveness in watermark recognition and comparison, with each network achieving over 80% accuracy for traced watermarks. EfficientNetB0 achieved 94.66%, VGG16 89.33%, ResNet50 86.67%, and InceptionV3 84%, while the prior art system gets 64,8%. These results conclude that these models are valuable tools in the field of watermark recognition and comparison.

A watermark image is a scan of a historical paper document that contains a watermark, which is a motif embedded in the paper that provides valuable information on the origins of a document. Developing tools to automatically identify watermarks can make this information more accessible to researchers. This paper focuses on one specific binarization technique, thresholding. Thresholding selects a threshold value, which is used to turn an image binary such that one color represents foreground and the other represents background. Ideally, binarization isolates the watermark’s shape by representing it as foreground, and removes unwanted information. This research compares the effectiveness of different thresholding techniques when applied to watermark images. Eight algorithms are selected from the literature, and a novel algorithm is proposed that seeks to improve on the other algorithms when applied to watermarks. The nine total algorithms are evaluated quantitatively on synthetic data, and qualitatively through a survey where participants select which algorithm appears best and rate it. The results show that there is no clear algorithm which works best for all images, however a logical adaptive approach may work marginally better than other approaches. Additionally, the presented algorithms do not adequately remove non-watermark information from the images. Further research should be conducted to analyze different binarization techniques in this context.

A curve is a continuously bending line with no angles that can be found anywhere in the real world, forming shapes and outlines. They are also the building blocks of historic watermarks, imprinted images on paper that may be used to identify its manufacturers. Their shapes consist of curves as bent wires are used in their production process. Often, the processing of scans of those curves may introduce gaps or a degraded quality which could be corrected by reconstructing the curves in those gaps. Curve reconstruction is a fundamental problem with many research applications, one of which is the reconstruction of curves for binarised scans of historic watermarks. In this paper, a data generation approach is proposed for the simulation of the watermark curves domain through singular automatically generated curves and human-drawn sketches which are then used along binarised watermark scans. I propose a hybrid method combining machine-learning and analytical approaches for curve reconstruction, aiming to leverage their advantages together. The method is compared to its components separately. Quantitative results against them demonstrate the superiority of the pure machine learning approach, as well as the need for more research into potentially better analytical components and a more realistic domain simulation.

Watermarks have an essential role in identifying the origins and age of specific documents. However, this is often a laborious process. One of the main issues in automatic watermark segmentation is the presence of text that obstructs it, making it difficult to properly reconstruct a watermark. Image processing and machine learning techniques face limitations, requiring time, training of data, or manual parameter selection. This research introduces a new method using wavelets transform to locate and remove text from a watermarked image, while preserving the underlying watermark. This method manages to outperform classic image processing techniques for the case when text is thicker than the watermark outline.

In the task of music style transfer, the symbolic music representation based on Musical Instrument Digital Interface (MIDI) files has always been a popular research medium. By using such representation, some mature models for image style transfer can also be applied to this scenario, such as Cycle-consistent Generative Adversarial Networks (CycleGAN). However, this MIDI-based data representation is not suitable for guitar music because it does not support unique expressive information of guitar playing, such as bending, sliding, or other playing techniques. DadaGP, a dataset made up of guitar-specific format files (tablatures) and their rendered text-like tokens, enables us to perform symbolic guitar music style transfer leveraging expressive guitar playing information, and to produce playable guitar tablatures. We first adopt K-hot encoding to transform the task from sequence generation to binary classification of multiple variables, and use top-$k$ sampling to reproduce sequences from output K-hot vectors. We then propose a novel model we call CycleGMT, a CycleGAN-based model for symbolic guitar music style transfer. Finally, to mitigate the severe sparsity in the data and its resulting content loss, we adopt a skip connection between the input and output of the model, successfully achieving style-transferred music whose quality being competitive with human-composed remixes, while the musical complexity of the style-transferred music can be controlled by adjusting the value of $k$ in top-$k$ sampling.

In today’s world, accurate location sensing is impossible to think away. One of the most prominent and most used techniques for determining location is GPS. In the outside world, GPS is capable of pinpointing a location with only a few meters error. But inside buildings, GPS often fails to deliver the same accuracy. In this paper, a relatively new technique will be presented to solve this problem using acoustic location sensing where a smartphone emits inaudible chirps and records the result. Specifically, this paper will cover what kind of data is needed to train the deep model that will solve this problem.

The work presented in this thesis investigates the creation of virtual sound sources in a room equipped with a limited number of loudspeakers. This limited number of loudspeakers is typical for consumer loudspeaker systems. Ideally, these systems can provide a listening experience in which localisation cues are completely present. However, in current systems, this is not the case. The limited number of loudspeakers make it impractical to produce a physically accurate sound field. A possible solution is to create a perceptually accurate sound field instead. In this work, a step towards an algorithm which can do so is presented. The developed algorithm requires knowledge of the listener placement, the room, and the loudspeaker placement. Spatial weighting is used to construct a region in which the acoustic en-ergy should be large and a region in which the acoustic energy should be limited. The loudspeaker playback signals are obtained by maximising the energy ratio between these regions, while at the same time ensuring that the perceptual difference between the received audio and reference audio remains limited. The algorithm employs a convex optimisation problem to facilitate efficiently solving for the playback signals. Concretely, six convex optimisation problems are proposed with somewhat increasing complexity and different weighting matrices. For each of the optimisation problems, the proposed algorithm is compared against a simple amplitude panning algorithm and a nearest neighbour algorithm. It is found that none of the considered algorithms is clearly preferred over the others in terms of the considered evaluation metrics. A major limitation of the presented work is that the evaluation metrics do not explicitly test for localisation accuracy. In future work, this should be investigated by including subjective tests.

Over the past decade, spatial audio awareness evolved into an in-demand feature in audio entertainment. The addition of sound source locations to, for instance, movies or music adds a level of auditory envelopment and spatial awareness to the audio experience. Expensive setups present in, for instance, cinema's, are able to create this envelopment by means of a large set of loudspeakers with which the desired sound fields are created. Creating this spatial envelopment in practical consumer living rooms or home cinema setups proves to be a more challenging task due to the impractical amount of loudspeakers required. To create the audio envelopment with a small amount of loudspeakers, crosstalk cancellation can be used. Crosstalk cancellation as posed in literature is, however, not robust enough to be used in practical appliances. The main cause of these bad characteristics is the objective cost function it optimizes which results in an ill-posed problem. In this thesis, the crosstalk cancellation problem is relaxed by aiming for perceptually sufficient results instead of aiming for objectively optimal results. The human auditory system has its limitations in both the perception of audio and localization of audio sources. Exploiting the limitations of the auditory system generates mathematical freedom that can be used to construct a more robust and stable crosstalk cancellation algorithm. This thesis provides the first steps towards the incorporation of audio perception in the domain of crosstalk cancellation and audio envelopment with a limited amount of loudspeakers.

The placement of a subwoofer has significant impact on the quality of its sound reproduction. This paper present a method to optimise subwoofer placement in a room. Resonances caused by room boundaries (called "room modes") cause localised peaks and lows for specific lower frequencies. The goal is to minimize the simulated frequency response over a listener region. To simulate the frequency response of a subwoofer location, a Finite-Difference Time-Domain method is applied. The results show that this method correlates to both a measured frequency response and an analytical model. The optimal location predicted with the simulation could not flatten every room mode, but had decreased variation for a number of resonant frequencies.

A novel approach for determining the optimal location of a sound source within an acoustic environment is proposed. This approach involves the application of Importance Sampling to improve the efficiency of the existing method of acoustic ray-tracing for finding the frequency response at various listening locations. The results of this study do not conclusively demonstrate the superiority of the proposed method over existing techniques, and further research is necessary to determine its viability.

Technology is playing an increasingly important role in education, partially thanks to emerging teaching methods such as hybrid education. Smart classrooms equipped with technology can make the lives of the educators and students easier, and aid in the switch to hybrid education. In this work we propose an IoT based smart classroom framework to create a base for smart classroom applications. Using this framework, multiple applications are manifested in the form of prototypes. One prototype introduces wireless sensor readout capabilities. The other prototype focuses on assisting the lecturer in a hybrid teaching setup. Using the prototype, the teacher is able to control whether online students are able to hear the students who are physically present and vice versa. The prototype also includes a way of indicating a question from an online student.