Searcharxiv⌕ Search

arXiv subjects

Huan Zhou

Publications and source records attributed to Huan Zhou.

At least 55 records · Page 3Linked to original sources

Enhancing Lip Reading with Multi-Scale Video and Multi-Encoder

Automatic lip-reading (ALR) aims to automatically transcribe spoken content from a speaker's silent lip motion captured in video. Current mainstream lip-reading approaches only use a single visual encoder to model input videos of a single scale. In this paper, we propose to enhance lip-reading by incorporating multi-scale video data and multi-encoder. Specifically, we first propose a novel multi-scale lip motion extraction algorithm based on the size of the speaker's face and an Enhanced ResNet3D visual front-end (VFE) to extract lip features at different scales. For the multi-encoder, in addition to the mainstream Transformer and Conformer, we also incorporate the recently proposed Branchformer and E-Branchformer as visual encoders. In the experiments, we explore the influence of different video data scales and encoders on ALR system performance and fuse the texts transcribed by all ALR systems using recognizer output voting error reduction (ROVER). Finally, our proposed approach placed second in the ICME 2024 ChatCLR Challenge Task 2, with a 21.52% reduction in character error rate (CER) compared to the official baseline on the evaluation set.

cs.CV↗

Exploring Supermassive Compact Dark Matter with the Millilensing Effect of Gamma-Ray Bursts

Gravitational lensing effect is one of most significant observational probes to investigate compact dark matter/objects over a wide mass range. In this work, we first propose to derive the population information and the abundance of supermassive compact dark matter in the mass range $\sim10^5-10^7~M_{\odot}$ from 6 millilensed gamma-ray burst (GRB) candidates in 3000 Fermi GRB events using the hierarchical Bayesian inference method. We obtain that, for the mass range $\sim10^5-10^7~M_{\odot}$, the abundance of supermassive compact dark matter is $f_{\rm CO}=10^{-1.60}$ in the log-normal mass distribution scenario. This result is in obvious tension with some other observational constraints, e.g. ultra-faint dwarfs and dynamical friction. However, it also was argued that there is only one system in these 6 candidates has been identified as lensed GRB event with fairly high confidence. In this case, the tension would be significantly alleviated. Therefore, it would be an interesting clue for both the millilensed GRB identification and the formation mechanism of supermassive compact dark matter.

astro-ph.HE↗

A method for constructing graphs with the same resistance spectrum

Let $G=(V(G),E(G))$ be a graph with vertex set $V(G)$ and edge set $E(G)$. The resistance distance $R_G(x,y)$ between two vertices $x,y$ of $G$ is defined to be the effective resistance between the two vertices in the corresponding electrical network in which each edge of $G$ is replaced by a unit resistor. The resistance spectrum $\mathrm{RS}(G)$ of a graph $G$ is the multiset of the resistance distances of all pairs of vertices in the graph. This paper presents a method for constructing graphs with the same resistance spectrum. It is obtained that for any positive integer $k$, there exist at least $2^k$ graphs with the same resistance spectrum. Furthermore, it is shown that for $n \geq 10$, there are at least $2(n-9) p(n-9)$ pairs of graphs of order $n$ with the same resistance spectrum, where $p(n-9)$ is the number of partitions of the integer $n-9$.

math.CO↗

Exploiting Low-level Representations for Ultra-Fast Road Segmentation

Achieving real-time and accuracy on embedded platforms has always been the pursuit of road segmentation methods. To this end, they have proposed many lightweight networks. However, they ignore the fact that roads are "stuff" (background or environmental elements) rather than "things" (specific identifiable objects), which inspires us to explore the feasibility of representing roads with low-level instead of high-level features. Surprisingly, we find that the primary stage of mainstream network models is sufficient to represent most pixels of the road for segmentation. Motivated by this, we propose a Low-level Feature Dominated Road Segmentation network (LFD-RoadSeg). Specifically, LFD-RoadSeg employs a bilateral structure. The spatial detail branch is firstly designed to extract low-level feature representation for the road by the first stage of ResNet-18. To suppress texture-less regions mistaken as the road in the low-level feature, the context semantic branch is then designed to extract the context feature in a fast manner. To this end, in the second branch, we asymmetrically downsample the input image and design an aggregation module to achieve comparable receptive fields to the third stage of ResNet-18 but with less time consumption. Finally, to segment the road from the low-level feature, a selective fusion module is proposed to calculate pixel-wise attention between the low-level representation and context feature, and suppress the non-road low-level response by this attention. On KITTI-Road, LFD-RoadSeg achieves a maximum F1-measure (MaxF) of 95.21% and an average precision of 93.71%, while reaching 238 FPS on a single TITAN Xp and 54 FPS on a Jetson TX2, all with a compact model size of just 936k parameters. The source code is available at https://github.com/zhouhuan-hust/LFD-RoadSeg.

cs.CV↗

Exploring primordial curvature perturbation on small scales with the lensing effect of fast radio bursts

Cosmological observations, e.g., cosmic microwave background, have precisely measured the spectrum of primordial curvature perturbation on larger scales, but smaller scales are still poorly constrained. Since primordial black holes (PBHs) could form in the very early Universe through the gravitational collapse of primordial density perturbations, constrains on the PBH could encodes much information on primordial fluctuations. In this work, we first derive a simple formula for lensing effect to apply PBH constraints with the monochromatic mass distribution to an extended mass distribution. Then, we investigate the latest fast radio burst observations with this relationship to constrain two kinds of primordial curvature perturbation models on the small scales. It suggests that, from the null search result of lensed fast radio burst in currently available observations, the amplitude of primordial curvature perturbation should be less than $8\times 10^{-2}$ at the scale region of $10^5-10^6~\rm Mpc^{-1}$. This corresponds to an interesting mass range relating to binary black holes detected by LIGO-Virgo-KAGRA and future Einstein Telescope or Cosmic Explorer.

astro-ph.CO↗

The c-differential properties of a class of power functions

Power functions with low $c$-differential uniformity have been widely studied not only because of their strong resistance to multiplicative differential attacks, but also low implementation cost in hardware. Furthermore, the $c$-differential spectrum of a function gives a more precise characterization of its $c$-differential properties. Let $f(x)=x^{\frac{p^n+3}{2}}$ be a power function over the finite field $\mathbb{F}_{p^{n}}$, where $p\neq3$ is an odd prime and $n$ is a positive integer. In this paper, for all primes $p\neq3$, by investigating certain character sums with regard to elliptic curves and computing the number of solutions of a system of equations over $\mathbb{F}_{p^{n}}$, we determine explicitly the $(-1)$-differential spectrum of $f$ with a unified approach. We show that if $p^n \equiv 3 \pmod 4$, then $f$ is a differentially $(-1,3)$-uniform function except for $p^n\in\{7,19,23\}$ where $f$ is an APcN function, and if $p^n \equiv 1 \pmod 4$, the $(-1)$-differential uniformity of $f$ is equal to $4$. In addition, an upper bound of the $c$-differential uniformity of $f$ is also given.

cs.IT↗

BA-MoE: Boundary-Aware Mixture-of-Experts Adapter for Code-Switching Speech Recognition

Mixture-of-experts based models, which use language experts to extract language-specific representations effectively, have been well applied in code-switching automatic speech recognition. However, there is still substantial space to improve as similar pronunciation across languages may result in ineffective multi-language modeling and inaccurate language boundary estimation. To eliminate these drawbacks, we propose a cross-layer language adapter and a boundary-aware training method, namely Boundary-Aware Mixture-of-Experts (BA-MoE). Specifically, we introduce language-specific adapters to separate language-specific representations and a unified gating layer to fuse representations within each encoder layer. Second, we compute language adaptation loss of the mean output of each language-specific adapter to improve the adapter module's language-specific representation learning. Besides, we utilize a boundary-aware predictor to learn boundary representations for dealing with language boundary confusion. Our approach achieves significant performance improvement, reducing the mixture error rate by 16.55\% compared to the baseline on the ASRU 2019 Mandarin-English code-switching challenge dataset.

cs.SD↗

Weak degeneracy of planar graphs and locally planar graphs

Weak degeneracy is a variation of degeneracy which shares many nice properties of degeneracy. In particular, if a graph $G$ is weakly $d$-degenerate, then for any $(d + 1)$-list assignment $L$ of $G$, one can construct an $L$-coloring of $G$ by a modified greedy coloring algorithm. It is known that planar graphs of girth 5 are 3-choosable and locally planar graphs are 5-choosable. This paper strengthens these results and proves that planar graphs of girth 5 are weakly 2-degenerate and locally planar graphs are weakly 4-degenerate.

math.CO↗

X-SepFormer: End-to-end Speaker Extraction Network with Explicit Optimization on Speaker Confusion

Target speech extraction (TSE) systems are designed to extract target speech from a multi-talker mixture. The popular training objective for most prior TSE networks is to enhance reconstruction performance of extracted speech waveform. However, it has been reported that a TSE system delivers high reconstruction performance may still suffer low-quality experience problems in practice. One such experience problem is wrong speaker extraction (called speaker confusion, SC), which leads to strong negative experience and hampers effective conversations. To mitigate the imperative SC issue, we reformulate the training objective and propose two novel loss schemes that explore the metric of reconstruction improvement performance defined at small chunk-level and leverage the metric associated distribution information. Both loss schemes aim to encourage a TSE network to pay attention to those SC chunks based on the said distribution information. On this basis, we present X-SepFormer, an end-to-end TSE model with proposed loss schemes and a backbone of SepFormer. Experimental results on the benchmark WSJ0-2mix dataset validate the effectiveness of our proposals, showing consistent improvements on SC errors (by 14.8% relative). Moreover, with SI-SDRi of 19.4 dB and PESQ of 3.81, our best system significantly outperforms the current SOTA systems and offers the top TSE results reported till date on the WSJ0-2mix.

eess.AS↗

CGI-Stereo: Accurate and Real-Time Stereo Matching via Context and Geometry Interaction

In this paper, we propose CGI-Stereo, a novel neural network architecture that can concurrently achieve real-time performance, competitive accuracy, and strong generalization ability. The core of our CGI-Stereo is a Context and Geometry Fusion (CGF) block which adaptively fuses context and geometry information for more effective cost aggregation and meanwhile provides feedback to feature learning to guide more effective contextual feature extraction. The proposed CGF can be easily embedded into many existing stereo matching networks, such as PSMNet, GwcNet and ACVNet. The resulting networks show a significant improvement in accuracy. Specially, the model which incorporates our CGF with ACVNet ranks $1^{st}$ on the KITTI 2012 and 2015 leaderboards among all the published methods. We further propose an informative and concise cost volume, named Attention Feature Volume (AFV), which exploits a correlation volume as attention weights to filter a feature volume. Based on CGF and AFV, the proposed CGI-Stereo outperforms all other published real-time methods on KITTI benchmarks and shows better generalization ability than other real-time methods. Code is available at https://github.com/gangweiX/CGI-Stereo.

cs.CV↗

Towards a reliable reconstruction of the power spectrum of primordial curvature perturbation on small scales from GWTC-3

Primordial black holes (PBHs) can be both candidates of dark matter and progenitors of binary black holes (BBHs) detected by the LIGO-Virgo-KAGRA collaboration. Since PBHs could form in the very early Universe through the gravitational collapse of primordial density perturbations, the population of BBHs detected by gravitational waves encodes much information on primordial curvature perturbation. In this work, we take a reliable and systematic approach to reconstruct the power spectrum of the primordial curvature perturbation from GWTC-3, under the hierarchical Bayesian inference framework, by accounting for the measurement uncertainties and selection effects. In addition to just considering the single PBH population model, we also report the results considering the multi-population model, i.e., the mixed PBH and astrophysical black hole binaries model. We find that the maximum amplitude of the reconstructed power spectrum of primordial curvature perturbation can be $\sim2.5\times10^{-2}$ at $\mathcal{O}(10^{5})~\rm Mpc^{-1}$ scales, which is consistent with the PBH formation scenario from inflation at small scales.

astro-ph.CO↗

Improving Target Speaker Extraction with Sparse LDA-transformed Speaker Embeddings

As a practical alternative of speech separation, target speaker extraction (TSE) aims to extract the speech from the desired speaker using additional speaker cue extracted from the speaker. Its main challenge lies in how to properly extract and leverage the speaker cue to benefit the extracted speech quality. The cue extraction method adopted in majority existing TSE studies is to directly utilize discriminative speaker embedding, which is extracted from the pre-trained models for speaker verification. Although the high speaker discriminability is a most desirable property for speaker verification task, we argue that it may be too sophisticated for TSE. In this study, we propose that a simplified speaker cue with clear class separability might be preferred for TSE. To verify our proposal, we introduce several forms of speaker cues, including naive speaker embedding (such as, x-vector and xi-vector) and new speaker embeddings produced from sparse LDA-transform. Corresponding TSE models are built by integrating these speaker cues with SepFormer (one SOTA speech separation model). Performances of these TSE models are examined on the benchmark WSJ0-2mix dataset. Experimental results validate the effectiveness and generalizability of our proposal, showing up to 9.9% relative improvement in SI-SDRi. Moreover, with SI-SDRi of 19.4 dB and PESQ of 3.78, our best TSE system significantly outperforms the current SOTA systems and offers the top TSE results reported till date on the WSJ0-2mix.

cs.SD↗

Possible discrimination of black hole origins from the lensing rate of DECIGO and B-DECIGO sources

In this paper, we forecast the expected detection rates and redshift distributions of gravitationally lensed gravitational waves (GWs) from three different mass distributions of primordial black holes (PBHs) and two stellar formation models of astrophysical black holes (ABHs) in the context of DECi-hertz Interferometer Gravitational wave Observatory (DECIGO) and it's smaller scale version B-DECIGO. It suggests that DECIGO will be able to detect $10^4-10^5$ GW signals from such binary black holes (BBHs) each year and the event rate distributions for PBHs will differ from those for ABHs due to their different merger rate with respect to redshift. The large number of event rates make $5-100$ detections of lensed GW signals being possible. After considering the gravitational lensing effect, the difference between the detection rates and distributions for PBHs and ABHs will be more significant. Therefore, this can be served as a complementary method to distinguish PBHs from ABHs.

astro-ph.CO↗

Containerisation for High Performance Computing Systems: Survey and Prospects

Containers improve the efficiency in application deployment and thus have been widely utilised on Cloud and lately in High Performance Computing (HPC) environments. Containers encapsulate complex programs with their dependencies in isolated environments making applications more compatible and portable. Often HPC systems have higher security levels compared to Cloud systems, which restrict users' ability to customise environments. Therefore, containers on HPC need to include a heavy package of libraries making their size relatively large. These libraries usually are specifically optimised for the hardware, which compromises portability of containers. Per contra, a Cloud container has smaller volume and is more portable. Furthermore, containers would benefit from orchestrators that facilitate deployment and management of containers at a large scale. Cloud systems in practice usually incorporate sophisticated container orchestration mechanisms as opposed to HPC systems. Nevertheless, some solutions to enable container orchestration on HPC systems have been proposed in state of the art. This paper gives a survey and taxonomy of efforts in both containerisation and its orchestration strategies on HPC systems. It highlights differences thereof between Cloud and HPC. Lastly, challenges are discussed and the potentials for research and engineering are envisioned.

cs.DC↗

Constraints on compact dark matter from lensing of gravitational waves for the third-generation gravitational wave detector

Since the first gravitational wave (GW) event from binary black hole (BBH) was detected by LIGO-Virgo, GWs have become a useful probe on astrophysics and cosmology. If compact dark matter (DM) objects e.g. primordial black holes, contribute a significant fraction of dark matter at wide mass range, they will cause microlensing in the GW signals with long wavelengths that are distinct from the lensing effects of electromagnetic signals from astrophysical objects. In this paper, we apply the lensing effect of GW from BBH to derive constraints on the abundance of compact DM for the Cosmic Explorer, a third-generation ground-based GW detector. We firstly consider two channels of formation of BBH that contribute to low and high redshift GW sources, including the astrophysical origin BBH scenario, and the primordial origin BBH scenario. Secondly, comparing with the method of optical depth, we use the Bayesian analysis to derive constraints on the abundance of compact DM with different mass function of lens taken into consideration. For a null search with $1000$ detected GW events of BBH, we find that the abundance of compact DM could be constrained to $\lesssim0.1\%$ in the mass range $\geq500~M_{\odot}$ at $68\%$ confidence level. In addition, if a GW event lensed by a compact DM object with $M_{\rm l}\in[100~M_{\odot},300~M_{\odot}]$ is detected in $100$ detected GW events of BBH, we can derive that the estimation of the abundance of compact DM is from $2.3\%$ to $25.2\%$ in this mass range with the Bayesian analysis.

astro-ph.CO↗

Speech Enhancement with Perceptually-motivated Optimization and Dual Transformations

To address the monaural speech enhancement problem, numerous research studies have been conducted to enhance speech via operations either in time-domain on the inner-domain learned from the speech mixture or in time--frequency domain on the fixed full-band short time Fourier transform (STFT) spectrograms. Very recently, a few studies on sub-band based speech enhancement have been proposed. By enhancing speech via operations on sub-band spectrograms, those studies demonstrated competitive performances on the benchmark dataset of DNS2020. Despite attractive, this new research direction has not been fully explored and there is still room for improvement. As such, in this study, we delve into the latest research direction and propose a sub-band based speech enhancement system with perceptually-motivated optimization and dual transformations, called PT-FSE. Specially, our proposed PT-FSE model improves its backbone, a full-band and sub-band fusion model, by three efforts. First, we design a frequency transformation module that aims to strengthen the global frequency correlation. Then a temporal transformation is introduced to capture long range temporal contexts. Lastly, a novel loss, with leverage of properties of human auditory perception, is proposed to facilitate the model to focus on low frequency enhancement. To validate the effectiveness of our proposed model, extensive experiments are conducted on the DNS2020 dataset. Experimental results show that our PT-FSE system achieves substantial improvements over its backbone, but also outperforms the current state-of-the-art while being 27\% smaller than the SOTA. With average NB-PESQ of 3.57 on the benchmark dataset, our system offers the best speech enhancement results reported till date.

cs.SD↗

Joint Speech Activity and Overlap Detection with Multi-Exit Architecture

Overlapped speech detection (OSD) is critical for speech applications in scenario of multi-party conversion. Despite numerous research efforts and progresses, comparing with speech activity detection (VAD), OSD remains an open challenge and its overall performance is far from satisfactory. The majority of prior research typically formulates the OSD problem as a standard classification problem, to identify speech with binary (OSD) or three-class label (joint VAD and OSD) at frame level. In contrast to the mainstream, this study investigates the joint VAD and OSD task from a new perspective. In particular, we propose to extend traditional classification network with multi-exit architecture. Such an architecture empowers our system with unique capability to identify class using either low-level features from early exits or high-level features from last exit. In addition, two training schemes, knowledge distillation and dense connection, are adopted to further boost our system performance. Experimental results on benchmark datasets (AMI and DIHARD-III) validated the effectiveness and generality of our proposed system. Our ablations further reveal the complementary contribution of proposed schemes. With $F_1$ score of 0.792 on AMI and 0.625 on DIHARD-III, our proposed system outperforms several top performing models on these datasets, but also surpasses the current state-of-the-art by large margins across both datasets. Besides the performance benefit, our proposed system offers another appealing potential for quality-complexity trade-offs, which is highly preferred for efficient OSD deployment.

cs.SD↗

Performance Analysis of Fog-Aided D2D Networks with Multicast-Based Opportunistic Content Delivery

In this paper, we develop a comprehensive and tractable analytical framework based on stochastic geometry to evaluate the performance of large-scale fog-aided device-to-device (F-D2D) networks with opportunistic content multicasting. As a part of the analysis, to resolve the contentions of file requests from the cache-incapable conventional user equipments (C-UEs), two simple yet typical candidate file selection schemes for cache-enabled fog user equipments (F-UEs), namely the random file selection (RFS) scheme and the most requested file selection (MRFS) scheme, are considered. Further, to suppress the harmful interference among the concurrent transmissions of F-UEs, a multicast-based opportunistic content delivery strategy is proposed by exploring the idea of opportunistic spectrum access (OSA). Assuming decentralized probabilistic caching, we first derive the activation probability of the F-UEs. Then, by adopting an appropriate approximation, the cache-hit probability, the coverage probability, and thereby the successful content delivery probability (SCDP) of the F-D2D network are evaluated. We also develop an iterative algorithm based on the gradient projection method to obtain a suboptimal caching policy for the maximization of SCDP. Extensive simulation and numerical results are presented to verify our analysis and demonstrate the superior performance of the proposed multicast-based opportunistic content delivery strategy.

cs.IT↗