Search

Pre-trained deep learning embeddings have consistently shown superior performance over handcrafted acoustic features in speech emotion recognition (SER). However, unlike acoustic features with clear physical meaning, these embeddings lack clear interpretability. Explaining these embeddings is crucial for building trust in healthcare and security applications and advancing the scientific understanding of the acoustic information that is encoded in them. This paper proposes a modified probing approach to explain deep learning embeddings in the SER space. We predict interpretable acoustic features (e.g., f0, loudness) from (i) the complete set of embeddings and (ii) a subset of the embedding dimensions identified as most important for predicting each emotion. If the subset of the most important dimensions better predicts a given emotion than all dimensions and also predicts specific acoustic features more accurately, we infer those acoustic features are important for the embedding model for the given task. We conducted experiments using the WavLM embeddings and eGeMAPS acoustic features as audio representations, applying our method to the RAVDESS and SAVEE emotional speech datasets. Based on this evaluation, we demonstrate that Energy, Frequency, Spectral, and Temporal categories of acoustic features provide diminishing information to SER in that order, demonstrating the utility of the probing classifier method to relate embeddings to interpretable acoustic features.

Ultrasonography is an essential tool in mid-pregnancy for assessing fetal development, appreciated for its non-invasive and real-time imaging capabilities. Yet, the interpretation of ultrasound images is often complicated by acoustic shadows, speckle noise, and other artifacts that obscure crucial diagnostic details. To address these challenges, our study presents a novel unsupervised anomaly detection framework specifically designed for fetal ultrasound imaging. This framework incorporates gestational age filtering, precise identification of fetal standard planes, and targeted segmentation of brain regions to enhance diagnostic accuracy. Furthermore, we introduce the use of denoising diffusion probabilistic models in this context, marking a significant innovation in detecting previously unrecognized anomalies. We rigorously evaluated the framework using various diffusion-based anomaly detection methods, noise types, and noise levels. Notably, AutoDDPM emerged as the most effective, achieving an area under the precision-recall curve of 79.8\% in detecting anomalies. This advancement holds promise for improving the tools available for nuanced and effective prenatal diagnostics., Comment: ASMUS 2024: 5th International Workshop on Advances in Simplifying Medical Ultrasound

Speaker identity plays a significant role in human communication and is being increasingly used in societal applications, many through advances in machine learning. Speaker identity perception is an essential cognitive phenomenon that can be broadly reduced to two main tasks: recognizing a voice or discriminating between voices. Several studies have attempted to identify acoustic correlates of identity perception to pinpoint salient parameters for such a task. Unlike other communicative social signals, most efforts have yielded inefficacious conclusions. Furthermore, current neurocognitive models of voice identity processing consider the bases of perception as acoustic dimensions such as fundamental frequency, harmonics-to-noise ratio, and formant dispersion. However, these findings do not account for naturalistic speech and within-speaker variability. Representational spaces of current self-supervised models have shown significant performance in various speech-related tasks. In this work, we demonstrate that self-supervised representations from different families (e.g., generative, contrastive, and predictive models) are significantly better for speaker identification over acoustic representations. We also show that such a speaker identification task can be used to better understand the nature of acoustic information representation in different layers of these powerful networks. By evaluating speaker identification accuracy across acoustic, phonemic, prosodic, and linguistic variants, we report similarity between model performance and human identity perception. We further examine these similarities by juxtaposing the encoding spaces of models and humans and challenging the use of distance metrics as a proxy for speaker proximity. Lastly, we show that some models can predict brain responses in Auditory and Language regions during naturalistic stimuli., Comment: Masters Thesis

Accurately predicting heart activity and other biological signals is crucial for diagnosis and monitoring. Given that speech is an outcome of multiple physiological systems, a significant body of work studied the acoustic correlates of heart activity. Recently, self-supervised models have excelled in speech-related tasks compared to traditional acoustic methods. However, the robustness of data-driven representations in predicting heart activity remained unexplored. In this study, we demonstrate that self-supervised speech models outperform acoustic features in predicting heart activity parameters. We also emphasize the impact of individual variability on model generalizability. These findings underscore the value of data-driven representations in such tasks and the need for more speech-based physiological data to mitigate speaker-related challenges., Comment: Accepted at Interspeech 2024

Magnonics is an emerging research field that addresses the use of spin waves (magnons), purely magnetic waves, for information transport and processing. Spin waves are a potential replacement for electric current in modern computational devices that would make them more compact and energy efficient. The field is yet little known, even among physicists. Additionally, with the development of new measuring techniques and computational physics, the obtained magnetic data becomes more complex, in some cases including 3D vector fields and time-resolution. This work presents an approach to the audio-visual representation of the spin waves and discusses its use as a tool for science communication exhibits and possible data analysis tool. The work also details an instance of such an exhibit presented at the annual international digital art exhibition Ars Electronica Festival in 2022., Comment: Accepted to The 29th International Conference on Auditory Display (ICAD 2024) conference proceedings

IPv6 is a fundamentally different Internet Protocol than IPv4, and IPv6-only networks cannot, by default, communicate with the IPv4 Internet. This lack of interoperability necessitates complex mechanisms for incremental deployment and bridging networks so that non-dual-stack systems can interact with the whole Internet. NAT64 is one such bridging mechanism by which a network allows IPv6-only clients to connect to the entire Internet, leveraging DNS to identify IPv4-only networks, inject IPv6 response addresses pointing to an internal gateway, and seamlessly translate connections. To date, our understanding of NAT64 deployments is limited; what little information exists is largely qualitative, taken from mailing lists and informal discussions. In this work, we present a first look at the active measurement of NAT64 deployment on the Internet focused on deployment prevalence, configuration, and security. We seek to measure NAT64 via two distinct large-scale measurements: 1) open resolvers on the Internet, and 2) client measurements from RIPE Atlas. For both datasets, we broadly find that despite substantial anecdotal reports of NAT64 deployment, measurable deployments are exceedingly sparse. While our measurements do not preclude the large-scale deployment of NAT64, they do point to substantial challenges in measuring deployments with our existing best-known methods. Finally, we also identify problems in NAT64 deployments, with gateways not following the RFC specification and also posing potential security risks.

Internet-wide scans are an important tool to evaluate the deployment of services. To enable large-scale application layer scans, a fast, stateless port scan (e.g., using ZMap) is often performed ahead of time to collect responsive targets. It is a common expectation that port scans on the entire IPv4 address space provide a relatively unbiased view as they cover the complete address space. Previous work, however, has found prefixes where all addresses share particular properties. In IPv6, aliased prefixes and fully responsive prefixes, i.e., prefixes where all addresses are responsive, are a well-known phenomenon. However, there is no such in-depth analysis for prefixes with these responsiveness patterns in IPv4. This paper delves into the underlying factors of this phenomenon in the context of IPv4 and evaluates port scans on a total of 161 ports (142 TCP & 19 UDP ports) from three different vantage points. To account for packet loss and other scanning artifacts, we propose the notion of a new category of prefixes, which we call highly responsive prefixes (HRPs). Our findings show that the share of HRPs can make up 70 % of responsive addresses on selected ports. Regarding specific ports, we observe that CDNs contribute to the largest fraction of HRPs on TCP/80 and TCP/443, while TCP proxies emerge as the primary cause of HRPs on other ports. Our analysis also reveals that application layer handshakes to targets outside HRPs are, depending on the chosen service, up to three times more likely to be successful compared to handshakes with targets located in HRPs. To improve future scanning campaigns conducted by the research community, we make our study's data publicly available and provide a tool for detecting HRPs. Furthermore, we propose an approach for a more efficient, ethical, and sustainable application layer target selection.

Privacy regulations have led to many websites showing cookie banners to their users. Usually, cookie banners present the user with the option to "accept" or "reject" cookies. Recently, a new form of paywall-like cookie banner has taken hold on the Web, giving users the option to either accept cookies (and consequently user tracking) or buy a paid subscription for a tracking-free website experience. In this paper, we perform the first completely automated analysis of cookiewalls, i.e., cookie banners acting as a paywall. We find cookiewalls on 0.6% of all queried 45k websites. Moreover, cookiewalls are deployed to a large degree on European websites, e.g., for Germany we see cookiewalls on 8.5% of top 1k websites. Additionally, websites using cookiewalls send 6.4 times more third-party cookies and 42 times more tracking cookies to visitors, compared to regular cookie banner websites. We also uncover two large Subscription Management Platforms used on hundreds of websites, which provide website operators with easy-to-setup cookiewall solutions. Finally, we publish tools, data, and code to foster reproducibility and further studies., Comment: Code and data available at https://bannerclick.github.io/

In this paper, we show that utilizing multiple protocols offers a unique opportunity to improve IP alias resolution and dual-stack inference substantially. Our key observation is that prevalent protocols, e.g., SSH and BGP, reply to unsolicited requests with a set of values that can be combined to form a unique device identifier. More importantly, this is possible by just completing the TCP hand-shake. Our empirical study shows that utilizing readily available scans and our active measurements can double the discovered IPv4 alias sets and more than 30x the dual-stack sets compared to the state-of-the-art techniques. We provide insights into our method's accuracy and performance compared to popular techniques.

The Internet architecture has facilitated a multi-party, distributed, and heterogeneous physical infrastructure where routers from different vendors connect and inter-operate via IP. Such vendor heterogeneity can have important security and policy implications. For example, a security vulnerability may be specific to a particular vendor and implementation, and thus will have a disproportionate impact on particular networks and paths if exploited. From a policy perspective, governments are now explicitly banning particular vendors, or have threatened to do so. Despite these critical issues, the composition of router vendors across the Internet remains largely opaque. Remotely identifying router vendors is challenging due to their strict security posture, indistinguishability due to code sharing across vendors, and noise due to vendor mergers. We make progress in overcoming these challenges by developing LFP, a tool that improves the coverage, accuracy, and efficiency of router fingerprinting as compared to the current state-of-the-art. We leverage LFP to characterize the degree of router vendor homogeneity within networks and the regional distribution of vendors. We then take a path-centric view and apply LFP to better understand the potential for correlated failures and fate-sharing. Finally, we perform a case study on inter- and intra-United States data paths to explore the feasibility to make vendor-based routing policy decisions, i.e., whether it is possible to avoid a particular vendor given the current infrastructure.

Users and businesses are increasingly deploying Internet of Things (IoT) devices at home, at work, and in factories. At the same time, we see an increase in the use of IPv6 for Internet connectivity. Even though the IoT ecosystem has been the focus of recent studies, there is no comprehensive analysis of IoT end-hosts in the IPv6 Internet to date. In this paper we perform an in-depth analysis of IPv6-reachable IoT hosts using active measurements. We run measurements targeting 530M IPv6 addresses on six popular IoT-related protocols. With 36.4K hosts in 156 countries we find 380x fewer IoT-speaking end-hosts compared to IPv4. Moreover, we conduct a security analysis for TLS-enabled IoT-speaking hosts identifying up to 57% untrusted certificates, with up to 32% being self-signed and 25% being expired. Finally, we plan to publish our measurement results, tools, and a website dashboard to foster further research in the area.

BGP exchanges reachability information in the form of prefixes, which are usually originated by a single Autonomous System (AS). If multiple ASes originate the same prefix, this is referred to as a Multiple Origin ASes (MOAS) prefix. One reason for MOAS prefixes are BGP prefix hijacks, which are mostly short-lived and have been studied extensively in the past years. In contrast to short-lived MOAS, long-lived MOAS have remained largely understudied. In this paper, we focus on long-lived MOAS prefixes and perform an in-depth study over six years. We identify around 24k long-lived MOAS prefixes in IPv4 and 1.4k in IPv6 being announced in January 2023. By analyzing the RPKI status we find that more than 40% of MOAS prefixes have all origins registered correctly, with only a minority of MOAS having invalid origins. Moreover, we find that the most prominent CIDR size of MOAS prefixes is /24 for IPv4 and /48 for IPv6, suggesting their use for fine-grained traffic steering. We attribute a considerable number of MOAS prefixes to mergers and acquisitions of companies. Additionally, more than 90% of MOAS prefixes are originated by two origin ASes, with the majority of detected origin AS relations being customer-provider. Finally, we identify that the majority of MOAS users are IT companies, and just 0.9% of IPv4 MOAS and 6.3% of IPv6 MOAS prefixes are used for anycast.

Internet measurements are a crucial foundation of IPv6-related research. Due to the infeasibility of full address space scans for IPv6 however, those measurements rely on collections of reliably responsive, unbiased addresses, as provided e.g., by the IPv6 Hitlist service. Although used for various use cases, the hitlist provides an unfiltered list of responsive addresses, the hosts behind which can come from a range of different networks and devices, such as web servers, customer-premises equipment (CPE) devices, and Internet infrastructure. In this paper, we demonstrate the importance of tailoring hitlists in accordance with the research goal in question. By using PeeringDB we classify hitlist addresses into six different network categories, uncovering that 42% of hitlist addresses are in ISP networks. Moreover, we show the different behavior of those addresses depending on their respective category, e.g., ISP addresses exhibiting a relatively low lifetime. Furthermore, we analyze different Target Generation Algorithms (TGAs), which are used to increase the coverage of IPv6 measurements by generating new responsive targets for scans. We evaluate their performance under various conditions and find generated addresses to show vastly differing responsiveness levels for different TGAs., Comment: Proceedings of the Network Traffic Measurement and Analysis Conference (TMA)

This paper addresses the important need for advanced techniques in continuously allocating workloads on shared infrastructures in data centers, a problem arising due to the growing popularity and scale of cloud computing. It particularly emphasizes the scarcity of research ensuring guaranteed capacity in capacity reservations during large-scale failures. To tackle these issues, the paper presents scalable solutions for resource management. It builds on the prior establishment of capacity reservation in cluster management systems and the two-level resource allocation problem addressed by the Resource Allowance System (RAS). Recognizing the limitations of Mixed Integer Linear Programming (MILP) for server assignment in a dynamic environment, this paper proposes the use of Deep Reinforcement Learning (DRL), which has been successful in achieving long-term optimal results for time-varying systems. A novel two-level design that utilizes a DRL-based algorithm is introduced to solve optimal server-to-reservation assignment, taking into account of fault tolerance, server movement minimization, and network affinity requirements due to the impracticality of directly applying DRL algorithms to large-scale instances with millions of decision variables. The paper explores the interconnection of these levels and the benefits of such an approach for achieving long-term optimal results in the context of large-scale cloud systems. We further show in the experiment section that our two-level DRL approach outperforms the MIP solver and heuristic approaches and exhibits significantly reduced computation time compared to the MIP solver. Specifically, our two-level DRL approach performs 15% better than the MIP solver on minimizing the overall cost. Also, it uses only 26 seconds to execute 30 rounds of decision making, while the MIP solver needs nearly an hour.

Flare frequency distributions represent a key approach to addressing one of the largest problems in solar and stellar physics: determining the mechanism that counter-intuitively heats coronae to temperatures that are orders of magnitude hotter than the corresponding photospheres. It is widely accepted that the magnetic field is responsible for the heating, but there are two competing mechanisms that could explain it: nanoflares or Alfv\'en waves. To date, neither can be directly observed. Nanoflares are, by definition, extremely small, but their aggregate energy release could represent a substantial heating mechanism, presuming they are sufficiently abundant. One way to test this presumption is via the flare frequency distribution, which describes how often flares of various energies occur. If the slope of the power law fitting the flare frequency distribution is above a critical threshold, $\alpha=2$ as established in prior literature, then there should be a sufficient abundance of nanoflares to explain coronal heating. We performed $>$600 case studies of solar flares, made possible by an unprecedented number of data analysts via three semesters of an undergraduate physics laboratory course. This allowed us to include two crucial, but nontrivial, analysis methods: pre-flare baseline subtraction and computation of the flare energy, which requires determining flare start and stop times. We aggregated the results of these analyses into a statistical study to determine that $\alpha = 1.63 \pm 0.03$. This is below the critical threshold, suggesting that Alfv\'en waves are an important driver of coronal heating., Comment: 1,002 authors, 14 pages, 4 figures, 3 tables, published by The Astrophysical Journal on 2023-05-09, volume 948, page 71