Generative models have shown remarkable performance in speech enhancement (SE), achieving superior perceptual quality over traditional discriminative approaches. However, existing generative SE approaches often overlook the risk of hallucination under severe noise, leading to incorrect spoken content or inconsistent speaker characteristics, which we term linguistic and acoustic hallucinations, respectively. We argue that linguistic hallucination stems from models' failure to constrain valid phonological structures and it is a more fundamental challenge. While language models (LMs) are well-suited for capturing the underlying speech structure through modeling the distribution of discrete tokens, existing approaches are limited in learning from noise-corrupted representations, which can lead to contaminated priors and hallucinations. To overcome these limitations, we propose the Phonologically Anchored Speech Enhancer (PASE), a generative SE framework that leverages the robust phonological prior embedded in the pre-trained WavLM model to mitigate hallucinations. First, we adapt WavLM into a denoising expert via representation distillation to clean its final-layer features. Guided by the model's intrinsic phonological prior, this process enables robust denoising while minimizing linguistic hallucinations. To further reduce acoustic hallucinations, we train the vocoder with a dual-stream representation: the high-level phonetic representation provides clean linguistic content, while a low-level acoustic representation retains speaker identity and prosody. Experimental results demonstrate that PASE not only surpasses state-of-the-art discriminative models in perceptual quality, but also significantly outperforms prior generative models with substantially lower linguistic and acoustic hallucinations.

Despite significant progress made in the last decade, deep neural network (DNN) based speech enhancement (SE) still faces the challenge of notable degradation in the quality of recovered speech under low signal-to-noise ratio (SNR) conditions. In this letter, we propose an SNR-progressive speech enhancement model with harmonic compensation for low-SNR SE. Reliable pitch estimation is obtained from the intermediate output, which has the benefit of retaining more speech components than the coarse estimate while possessing a significantly higher SNR than the input noisy speech. An effective harmonic compensation mechanism is introduced for better harmonic recovery. Extensive experiments demonstrate the advantage of our proposed model.

MetricGAN and its variations have been proven to be an effective wide-band speech enhancement model. In this paper, we expand it to full-band enhancement by combining our recently proposed learnable spectral dimension compression mapping strategy. The encoder-decoder structure with a time-frequency convolutional recurrent network is utilized as the generator. The proposed model is submitted to the ICASSP Signal Processing Grand Challenge: DNS-5 Challenge (2023). Without using the enrollment speech, it obtains a final score of 0.548 on Track-1 and 0.559 on Track-2.

The trustworthiness of vehicle messages is a major focus in intelligent transportation research. Existing studies focus on enhancing the accuracy of vehicle trustworthiness evaluation, overlooking that the transmission performance may affect the quality of vehicle messages, which are essential for implementing trustworthiness evaluation. This paper studies a novel vehicle message scheduling scheme to regulate vehicles’ transmissions so that incident messages for any part of the road can be reliably and accurately sent to a roadside unit. Through strategically scheduling vehicle transmissions to avoid interference between vehicles while guaranteeing sufficient numbers of vehicles transmitting their sensor data, vehicle messages can reliably yet timely arrive at Road Side Units (RSUs) without missing reporting an event on the road.

HTTPS refers to an application-specific implementation that runs HyperText Transfer Protocol (HTTP) on top of Secure Socket Layer (SSL) or Transport Layer Security (TLS). HTTPS is used to provide encrypted communication and secure identification of web servers and clients, for different purposes such as online banking and e-commerce. However, many HTTPS vulnerabilities have been disclosed in recent years. Although many studies have pointed out that these vulnerabilities can lead to serious consequences, domain administrators seem to ignore them. In this study, we evaluate the HTTPS security level of Alexa’s top 1 million domains from two perspectives. First, we explore which popular sites are still affected by those well-known security issues. Our results show that less than 0.1% of HTTPS-enabled servers in the measured domains are still vulnerable to known attacks including Rivest Cipher 4 (RC4), Compression Ratio Info-Leak Mass Exploitation (CRIME), Padding Oracle On Downgraded Legacy Encryption (POODLE), Factoring RSA Export Keys (FREAK), Logjam, and Decrypting Rivest–Shamir–Adleman (RSA) using Obsolete and Weakened eNcryption (DROWN). Second, we assess the security level of the digital certificates used by each measured HTTPS domain. Our results highlight that less than 0.52% domains use the expired certificate, 0.42% HTTPS certificates contain different hostnames, and 2.59% HTTPS domains use a self-signed certificate. The domains we investigate in our study cover 5 regions (including ARIN, RIPE NCC, APNIC, LACNIC, and AFRINIC) and 61 different categories such as online shopping websites, banking websites, educational websites, and government websites. Although our results show that the problem still exists, we find that changes have been taking place when HTTPS vulnerabilities were discovered. Through this three-year study, we found that more attention has been paid to the use and configuration of HTTPS. For example, more and more domains begin to enable the HTTPS protocol to ensure a secure communication channel between users and websites. From the first measurement, we observed that many domains are still using TLS 1.0 and 1.1, SSL 2.0, and SSL 3.0 protocols to support user clients that use outdated systems. As the previous studies revealed security risks of using these protocols, in the subsequent studies, we found that the majority of domains updated their TLS protocol on time. Our 2020 results suggest that most HTTPS domains use the TLS 1.2 protocol and show that some HTTPS domains are still vulnerable to the existing known attacks. As academics and industry professionals continue to disclose attacks against HTTPS and recommend the secure configuration of HTTPS, we found that the number of vulnerable domain is gradually decreasing every year.

Traditional address scanning attacks mainly rely on the naive 'brute forcing' approach, where the entire IPv4 address space is exhaustively searched by enumerating different possibilities. However, such an approach is inefficient for IPv6 due to its vast subnet size (i.e., 264). As a result, it is widely assumed that address scanning attacks are less feasible in IPv6 networks. In this paper, we evaluate new IPv6 reconnaissance techniques in real IPv6 networks and expose how to leverage the Domain Name System (DNS) for IPv6 network reconnaissance. We collected IPv6 addresses from 5 regions and 100,000 domains by exploiting DNS reverse zone and DNSSEC records. We propose a DNS Guard (DNSG) to efficiently detect DNS reconnaissance attacks in IPv6 networks. DNSG is a plug and play component that could be added to the existing infrastructure. We implement DNSG using Bro and Suricata. Our results demonstrate that DNSG could effectively block DNS reconnaissance attacks.

In the Public Key infrastructure (PKI) model, digital certificates play a vital role in securing online communication. Communicating parties exchange and validate these certificates, the validation fails if a certificate has been revoked. In this paper we propose the Certificate Revocation Guard (CRG) to efficiently check certificate revocation while minimising bandwidth, latency and storage overheads. CRG is based on OCSP, which caches the status of certificates locally. CRG could be installed on the user's machine, at the organisational proxy or even at the ISP level. Compared to a naive approach (where a client checks the revocation status of all certificates in the chain on every request), CRG decreases the bandwidth overheads and network latencies by 95%. Using CRG incurs 69% lower storage overheads compared to the CRL method. Our results demonstrate the effectiveness of our approach to improve certificate revocation.