AT&T 实验室

Affective computing aims to understand and model human emotions for computational systems. Within this field, speech emotion recognition (SER) focuses on predicting emotions conveyed through speech. While early SER systems relied on limited datasets and traditional machine learning models, recent deep learning approaches demand largescale, naturalistic emotional corpora. To address this need, we introduce the MSP-Conversation corpus: a dataset of more than 70 hours of conversational audio with time-continuous emotional annotations and detailed speaker diarizations. The time-continuous annotations capture the dynamic and contextdependent nature of emotional expression. The annotations in the corpus include fine-grained temporal traces of valence, arousal, and dominance. The audio data is sourced from publicly available podcasts and overlaps with a subset of the isolated speaking turns in the MSP-Podcast corpus to facilitate direct comparisons between annotation methods (i.e., in-context versus out-of-context annotations). The paper outlines the development of the corpus, annotation methodology, analyses of the annotations, and baseline SER experiments, establishing the MSP-Conversation corpus as a valuable resource for advancing research in dynamic SER in naturalistic settings.

Wireless backhaul is a cost-effective alternative to fiber in mobile networks. However, meeting rising reliability and capacity requirements over wireless point-to-point links remains challenging as the high-frequency links required for multi-Gbps transport are highly sensitive to precipitation. To quantify this effect, we develop an approach based on International Telecommunication Union (ITU) rain-attenuation models that predicts the capacity that wireless point-to-point links can guarantee at different reliability levels. Using frequency bands and system parameters representative of current deployments, we evaluate the impact of climate and link distance on guaranteed capacity. Our results show that in rainy regions, higher-frequency bands cannot deliver ultra-reliable capacity for typical link distances. Motivated by this, we assess two mitigation strategies: combining capacity from multiple bands and multi-hop relaying over shorter distances. We find that combining high and low bands can enable ultra-reliable operation, albeit with reduced capacities, while relaying improves guaranteed capacity at moderate reliability levels but cannot ensure ultra-reliability.

Managing fronthaul slices in 6G networks requires not only proactive bandwidth provisioning but also rapid adaptation to sudden and unpredictable traffic fluctuations. Although machine learning (ML) models have shown promise in forecasting regular traffic patterns, their performance often degrades under sudden demand surges, such as those caused by large-scale events. To address this challenge, we propose Z-PRISM (Zero-touch Proactive Resizer for Intelligent Slice Management), a data-driven solution that combines the Time-Series-Transformer (TST) and the AutoRegressive (AR) algorithm to adapt to abrupt changes in traffic patterns. Z-PRISM can predict short-term traffic dynamics and proactively reconfigure fronthaul slices in near real time. Simulation results demonstrate that Z-PRISM maintains more than 90% synchronization between O-RUs and BBUs during sudden traffic changes, outperforming ML-based alternatives that drop below 50%. The proposed solution aligns with the zero-touch management paradigm, offering a fast, resilient, and computationally efficient approach to adaptive slice management in dynamic 6G fronthaul environments.

Teaching and scientific experimentation with graphs require varied and precise instances, which are often laborious and impractical to generate manually. To address this, we present a graph generator designed for pedagogical and research applications that addresses the difficulties of manual creation. This artifact, named CG2E, consists of two modules: the Conventional (GC), which prioritizes strict adherence to user-specified parameters, and the Scale-Free (GSF), which generates graphs with power-law degree distributions. The validation of the CG2E involved 8,460 experimental tests spanning graphs with 100 to 1 million vertices, evaluating success rate, parametric accuracy, and statistical fit quality. Results show the GC module achieved 96.80% success and 100% accuracy as specified, while the GSF module achieved 100% success across all instances with a statistically significant power-law fit. This evidence indicates that the generator is a robust solution for educational and scientific purposes, enabling students, teachers, and researchers to work with diverse graphs and to obtain reproducible, parameter-specific instances for evaluating algorithms and methods.

This paper highlights results from a DOE-funded independent study to assess potential interference in the 6 GHz band from outdoor unlicensed Wi-Fi operations on Fixed Service (FS) microwave links used by the electric sector. The project conducts real-world experiments to evaluate interference risks and test the FCC-mandated Automated Frequency Coordination (AFC) system’s effectiveness. Field tests with a Southern Company FS link in Georgia revealed under-protection by AFC systems and additive spectrum interference not accounted for in current regulation. Results raise questions about assumptions in path loss models used in the AFC. The paper concludes with two recommendations on how to mitigate AFC under-protection of incumbents and proposes future experiments with other unlicensed device types.