Efficient time-domain speech separation using short encoded sequence network

With the growing use of short-video and audio data in personalized psychological assessment, multimodal information provides richer feature support for Big Five personality prediction tasks. However, effective fusion and modeling of multimodal data remain challenging. This paper proposes a multimodal Big Five prediction model based on an improved Vision-Mamba visual encoder and an S6-enhanced xLSTM temporal module. In the visual branch we retain Vision-Mamba’s bidirectional state-space module for global dependency modeling and innovatively append lightweight channel-wise 1D depth-separable convolution modules after layers 1, 7, 13 and 18 to enhance local texture consistency between adjacent patches, compensating for the global model’s limited sensitivity to local details. In the temporal branch we employ xLSTM for sequence modeling and extend it by introducing a selective state-space module, which improves long-range dependency capture efficiency without increasing computational complexity. For the audio branch we extract 40-dim MFCCs, time-frequency spectral and chroma features, map them via multi-layer fully connected networks to 128-dim embeddings. Finally, a feature-level fusion strategy concatenates visual and audio representations and feeds them to a multi-layer perceptron that outputs five personality scores. Experiments on the publicly available and widely used First Impressions V2 dataset [1] show that the VxL model achieves comparable prediction accuracy while significantly outperforming traditional multimodal Big Five models in training speed and GPU memory usage, validating the effectiveness of the proposed improvements.

The encoder-decoder architecture has been widely adopted by many handwritten mathematical expression recognition (HMER) models. However, these methods may struggle to accurately recognize expressions with complex structures or long sequences due to the lack of global information utilization. To address this issue, we propose a novel network based on the encoder-decoder framework, called the Structure and Counting Aware Network (SCAN), to enable global information awareness and achieve more precise recognition. Specifically, we introduce the Skeleton Shaping and Character Counting Module (SSCCM) to extract the skeleton structure of expressions and the frequency distribution of characters. These two types of information are integrated with the decoder output in the calibration module to refine the predictions. Experimental results demonstrate that SCAN significantly outperforms the existing state-of-the-art (SOTA) models in terms of expression recognition rate (ExpRate) on the CROHME 2014/2016/2019 and HME100K test sets. The code is publicly available at https://github.com/Kerston12138/SCAN.

In this study, we address a fundamental yet challenging problem on detection and classification of blurred regions in partially blurred images. we propose to learn a latent feature representation with stacked auto-encoder (SAE) network to perform blur region detection. Most previous approaches focus on extracting a few blur features in image gradient, Fourier domain, and data-driven local filters. We extract a latent high-level feature representation from such low-level features using the stacked auto-encoder network, thereby improve the accuracy of blur region classification. This high accuracy enables us to successfully separate the clear and blurred regions. Experimental results demonstrate that the proposed method significantly outperforms the state-of-the-arts methods in detecting and classifying blur regions in partially blurred images.

As the fruit-picking process is a labour-intensive and time-consuming task, the accurate and efficient recognition of apples during picking is of great significance for improving the overall efficiency of apple harvesting, reducing labour costs, and enhancing the quality of fruit picking. Although YOLOV5 algorithm can effectively detect apple status, its use on portable mobile devices still faces problems such as running lag. This paper is dedicated to the optimization of the C3-Light lightweight algorithm based on the widely used YOLOv5 framework for apple-picking recognition. The network structure of the C3-Light algorithm is redesigned. By introducing novel convolutional block arrangements and fine-tuning the hyperparameters related to the network layers, the model's feature extraction ability is enhanced while maintaining a relatively simple architecture. Through these improvements, the calls for hardware resources are remarkably reduced. Experimental results clearly demonstrate that the lightweight C3-Light model can maintain the original high-level accuracy. Specifically, it reduces GFLOPs by approximately 17% compared to the original model, which means a significant decrease in the computational complexity. Moreover, the GPU memory usage is decreased by 11%, indicating better memory utilization efficiency.

Accurate segmentation of medical images is crucial for disease diagnosis and understanding disease changes. Deep learning methods, utilizing encoder-decoder structures, have demonstrated cutting-edge performance in various medical image segmentation tasks. However, the pooling operation in the encoding stage results in feature loss, which makes the network lack the ability to fuse multi-scale information at different levels, hinders its effective perception of multi-scale information, and leads to poor segmentation performance. Drawing inspiration from the U-shaped network, this study introduces a multi-branch feature hybrid attention and adaptive receptive field network (MFHARFNet) for medical image segmentation. Building upon the encoder-decoder framework, we initially devise a multi-branch feature hybrid attention module (MFHAM) to seamlessly integrate feature maps of varying scales, capturing both fine-grained features and coarse-grained semantics across the entire scale. Furthermore, we redesign the skip connection to amalgamate feature information from different branches in the encoder stage and efficiently transmit it to the decoder, providing the decoder with global context feature maps at different levels. Finally, the adaptive receptive field (ARF) module is introduced in the decoder feature reconstruction stage to adapt and focus on related fields, ensuring the model’s adaptation to different segmentation target features, and achieving different weights for the output of different convolution kernels to improve segmentation performance. We comprehensively evaluate our method on medical image segmentation tasks, by using four public datasets across CT and MRI. Remarkably, MFHARFNet method consistently outperforms other state-of-the-art methods, exceeding UNet by 2.1%, 0.9%, 6.6% and 1.0% on Dice on ATLAS, LiTs, BraTs2019 and Spine and intervertebral disc datasets, respectively. In addition, MFHARFNet minimizes network parameters and computational complexity as much as possible. The source codes are in https://github.com/OneHundred99/MFHARFNet.

• VTPS Efficiency: The vane-type pipe separator (VTPS) is effective for compact, high-efficiency oil-water separation in constrained oilfield environments. • Advanced Modeling: A computational framework using the AIAD model within an Euler-Euler multiphase context is developed and experimentally validated. • Key Parameters Identified: Sensitivity analysis reveals guide vane angle and oil fraction as dominant factors affecting pressure drop and oil recovery. • Optimization Approach: Multi-objective optimization using SPEA2 and TOPSIS identifies the best trade-off between oil recovery and hydraulic loss. • Optimal Design Achieved: An optimal VTPS setup (25.4° vane angle, 3.57 m³/h flow rate, 18.4% oil fraction, 0.12 split ratio) achieves a 0.162 oil recovery factor at 2422 Pa pressure drop. Efficient oil-water separation is increasingly vital in mature oilfields, where enhanced water production challenges conventional downhole separation technologies. The vane-type pipe separator (VTPS) offers a promising solution due to its compact configuration and superior phase separation efficiency in constrained environments. However, the inherent complexities of swirling multiphase flow necessitate advanced modeling approaches for performance optimization. In this study, a comprehensive computational framework based on the Algebraic Interfacial Area Density (AIAD) model within an Euler-Euler multiphase context is developed and validated against benchmarked experimental data. Sensitivity analysis employing response surface methodology (RSM) and global-local indices quantifies the influence of key decision parameters, including guide vane installation angle, inlet flow rate, oil fraction, and split ratio, on separator performance. The study elucidates the parametric interplay affecting pressure drop and oil recovery, revealing guide vane angle and oil fraction as the dominant factors for hydraulic and separation performance, respectively. A multi-objective optimization is conducted using the Strength Pareto Evolutionary Algorithm 2 (SPEA2), with final selection by TOPSIS, to maximize oil recovery while minimizing hydraulic losses. The proposed methodology identifies an optimal configuration with a 25.4° vane angle, 3.57 m³/h flow rate, 18.4% oil fraction, and 0.12 split ratio, achieving a 0.162 oil recovery factor at a pressure drop of 2422 Pa. These findings provide actionable guidelines for VTPS design and operation, facilitating more efficient separation in oilfield applications and demonstrating the AIAD model's efficacy for industrial-scale multiphase flow optimization.

In this paper we introduce an advanced supervised training method for neural networks. It is based on Jacobian rank deficiency and it is formulated, in some sense, in the spirit of the Gauss-Newton algorithm. The Levenberg-Marquardt algorithm, as a modified Gauss-Newton, has been used successfully in solving nonlinear least squares problems including neural-network training. It outperforms (in terms of training accuracy, convergence properties, overall training time, etc.) the basic backpropagation and its variations with variable learning rate significantly, however, with higher computation and memory complexities within each iteration. The new method developed in this paper is aiming at improving convergence properties, while reducing the memory and computation complexities in supervised training of neural networks. Extensive simulation results are provided to demonstrate the superior performance of the new algorithm over the Levenberg-Marquardt algorithm.

Heat Exchange and Separation Efficiency in a Cluster of Gas-solid Separators in a Complex Cement Production Plant

High performance lamellar structured graphene oxide nanocomposite membranes via Fe3O4-coordinated phytic acid control of interlayer spacing for organic solvent nanofiltration (OSN)

Social media content can be used as a complementary source to the traditional methods for extracting and studying collective social attributes. This study focuses on the prediction of the occupational class for a public user profile. Our analysis is conducted on a new annotated corpus of Twitter users, their respective job titles, posted textual content and platform-related attributes. We frame our task as classification using latent feature representations such as word clusters and embeddings. The employed linear and, especially, non-linear methods can predict a user’s occupational class with strong accuracy for the coarsest level of a standard occupation taxonomy which includes nine classes. Combined with a qualitative assessment, the derived results confirm the feasibility of our approach in inferring a new user attribute that can be embedded in a multitude of downstream applications.

Separation of speech mixtures in noisy and reverberant environments remains a challenging task for state-of-the-art speech separation systems. Time-domain audio speech separation networks (TasNets) are among the most commonly used network architectures for this task. TasNet models have demonstrated strong performance on typical speech separation baselines where speech is not contaminated with noise. When additive or convolutive noise is present, performance of speech separation degrades significantly. TasNets are typically constructed of an encoder network, a mask estimation network and a decoder network. The design of these networks puts the majority of the onus for enhancing the signal on the mask estimation network when used without any pre-processing of the input data or post processing of the separation network output data. Use of multihead attention (MHA) is proposed in this work as an additional layer in the encoder and decoder to help the separation network attend to encoded features that are relevant to the target speakers and conversely suppress noisy disturbances in the encoded features. As shown in this work, incorporating MHA mechanisms into the encoder network in particular leads to a consistent performance improvement across numerous quality and intelligibility metrics on a variety of acoustic conditions using the WHAMR corpus, a data-set of noisy reverberant speech mixtures. The use of MHA is also investigated in the decoder network where it is demonstrated that smaller performance improvements are consistently gained within specific model configurations. The best performing MHA models yield a mean 0.6 dB scale invariant signal-to-distortion (SISDR) improvement on noisy reverberant mixtures over a baseline 1D convolution encoder. A mean 1 dB SISDR improvement is observed on clean speech mixtures.

In order to meet the demand of industrial production, an axial separator for oil-water separation is proposed. The separator uses the pattern of multi-stage separation, and it is divided into two chambers by two swirl impellers. The oil phase flows out through three Light Phase Outlet (LPO) at the center of the impeller hub, meanwhile the water phase flows out through Heavy Phase Outlet (HPO). An experimental platform was built to verify the feasibility of multi-stage separation. The Mixture model and Reynolds stress model was used to simulate the separation process, and the numerical results had a well agreement with experimental results. The effects of the diameter of cylindrical section, length of conical section and length of separation chamber on the separation performance were studied. The results show that: changing the diameter of cylindrical section has greatest impact on separation performance. The separation efficiency and pressure drop first increase and then decrease with increase of primary conical section length. With increase of second conical section length, the separation performance has been increasing. If it only looks at the relationship between pressure drop and separation efficiency, then there is a pressure drop to maximize the separation efficiency. Changing the diameter of cylindrical section can adjust the flow split LPOs. Increasing the length of second conical section can increase the split ratio of LPO3, but it has almost no effect on the total split ratio of LPOs. the value of structure is determined and the efficiency of separator exceeds 93 %.

Response surface method to study the effect of conical surface and vortex-finder lengths on de-foulant hydrocyclone with reflux ejector

Sulfur hexafluoride (SF6) is widely used in the power industry and significantly contributes to the greenhouse effect, necessitating the development of efficient materials for SF6 capture, particularly fluorine-containing materials. However, existing fluorine-containing materials often require complex monomers and high synthesis temperatures. Herein, we report the synthesis of a fluorine-functionalized carbazole-based nanoporous organic polymer (CNOP-7) at room temperature, using commercially available 4,4'-bis(9H-carbazole-9-yl)-1,1'-biphenyl and 1,1,1-trifluoroacetone. CNOP-7 contains 14.7% fluorine atoms and exhibits a high specific surface area of 1270 m2·g-1, demonstrating excellent SF6 adsorption and separation performance. The SF6/N2 selectivity of CNOP-7 reaches 107 at 273 K and 73 at 298 K. Furthermore, dynamic breakthrough experiments confirm that CNOP-7 can efficiently and repeatedly separate SF6 from SF6/N2 mixtures. Molecular simulations reveal the mechanism behind its efficient separation. This work offers fresh perspectives on the development and fabrication of adsorbents for efficient SF6 sequestration.

Cyclone separators are suitable to remove particles in extreme conditions with high pressure and temperature during the gas-solid separation process. The important performance of cyclone separators is determined by the separation efficiency and pressure drop. The inlet parameters including the number of inlets, inlet geometry and inlet gas velocity affects the performance of cyclone separators. In this paper, the experimental investigation has been conducted to determine the effects of dual inlet to the separation efficiency of cyclone separators. The cyclone separator was fabricated using the acrylic materials to accommodate the flow observation and trajectories inside the chamber. The experiments were conducted at 9, 11 and 13 m/s of inlet velocity using three particle sizes at 277.5, 42.5 and 625 um. Results showed that as the particle sizes increases, the separation efficiency also increases up to 94.5% for 625 um particles. The findings indicate that the dual inlet cyclone separators increases the performance of cyclone separators especially to remove fly ash in the biomass processing plant and other gas-solid process.