Search

Transformers have emerged as a prominent model framework for audio tagging (AT), boasting state-of-the-art (SOTA) performance on the widely-used Audioset dataset. However, their impressive performance often comes at the cost of high memory usage, slow inference speed, and considerable model delay, rendering them impractical for real-world AT applications. In this study, we introduce streaming audio transformers (SAT) that combine the vision transformer (ViT) architecture with Transformer-Xl-like chunk processing, enabling efficient processing of long-range audio signals. Our proposed SAT is benchmarked against other transformer-based SOTA methods, achieving significant improvements in terms of mean average precision (mAP) at a delay of 2s and 1s, while also exhibiting significantly lower memory usage and computational overhead. Checkpoints are publicly available https://github.com/RicherMans/SAT.

Previously, Target Speaker Extraction (TSE) has yielded outstanding performance in certain application scenarios for speech enhancement and source separation. However, obtaining auxiliary speaker-related information is still challenging in noisy environments with significant reverberation. inspired by the recently proposed distance-based sound separation, we propose the near sound (NS) extractor, which leverages distance information for TSE to reliably extract speaker information without requiring previous speaker enrolment, called speaker embedding self-enrollment (SESE). Full- & sub-band modeling is introduced to enhance our NS-Extractor's adaptability towards environments with significant reverberation. Experimental results on several cross-datasets demonstrate the effectiveness of our improvements and the excellent performance of our proposed NS-Extractor in different application scenarios., Accepted by InterSpeech2023

The currently most prominent algorithm to train keyword spotting (KWS) models with deep neural networks (DNNs) requires strong supervision i.e., precise knowledge of the spoken keyword location in time. Thus, most KWS approaches treat the presence of redundant data, such as noise, within their training set as an obstacle. A common training paradigm to deal with data redundancies is to use temporally weakly supervised learning, which only requires providing labels on a coarse scale. This study explores the limits of DNN training using temporally weak labeling with applications in KWS. We train a simple end-to-end classifier on the common Google Speech Commands dataset with increased difficulty by randomly appending and adding noise to the training dataset. Our results indicate that temporally weak labeling can achieve comparable results to strongly supervised baselines while having a less stringent labeling requirement. In the presence of noise, weakly supervised models are capable to localize and extract target keywords without explicit supervision, leading to a performance increase compared to strongly supervised approaches.

Large-scale audio tagging datasets inevitably contain imperfect labels, such as clip-wise annotated (temporally weak) tags with no exact on- and offsets, due to a high manual labeling cost. This work proposes pseudo strong labels (PSL), a simple label augmentation framework that enhances the supervision quality for large-scale weakly supervised audio tagging. A machine annotator is first trained on a large weakly supervised dataset, which then provides finer supervision for a student model. Using PSL we achieve an mAP of 35.95 balanced train subset of Audioset using a MobileNetV2 back-end, significantly outperforming approaches without PSL. An analysis is provided which reveals that PSL mitigates missing labels. Lastly, we show that models trained with PSL are also superior at generalizing to the Freesound datasets (FSD) than their weakly trained counterparts., Accepted by ICASSP 2022

We study the usability of pre-trained weakly supervised audio tagging (AT) models as feature extractors for general audio representations. We mainly analyze the feasibility of transferring those embeddings to other tasks within the speech and sound domains. Specifically, we benchmark weakly supervised pre-trained models (MobileNetV2 and EfficientNet-B0) against modern self-supervised learning methods (BYOL-A) as feature extractors. Fourteen downstream tasks are used for evaluation ranging from music instrument classification to language classification. Our results indicate that AT pre-trained models are an excellent transfer learning choice for music, event, and emotion recognition tasks. Further, finetuning AT models can also benefit speech-related tasks such as keyword spotting and intent classification., Comment: Odyssey 2022

In this paper, a method combining the orthogonal array design and the numerical simulation is proposed to optimize the geometry parameters of the solution blowing nozzles. The centerline velocity is used to evaluate the performance of the nozzle and the characteristics of airflow fields are calculated. Three geometry parameters of the nozzle: the protrusion length of needle, the diameter of needle and the diameter of nozzle are investigated. The results show that smaller needle diameter and larger nozzle diameter will result in a higher centerline velocity, which is beneficial to fiber attenuation, whereas the effect of needle protrusion length is insignificant. The optimal geometry parameters of the nozzle achieved in this study are that the protrusion length of needle of 5 mm, the diameter of needle of 0.8 mm, and the diameter of nozzle of 4 mm. Furthermore, chitosan/PEO nanofibers are manufactured and studied with different geometry nozzles. This work can provide a better understanding of the controllable fabrication of solution blown nanofibers.

The microstructure evolution of Al–Sr master alloy during continuous extrusion was investigated using X-ray diffractometer, scanning electron microscope and transmission electron microscope. Results indicate that the continuous extrusion process could change the Al4Sr particles of the alloy significantly in size and morphology. The as-cast needle-like Al4Sr particles are broken into small blocks in upsetting zone and crushed heavily in adhesion zone. Plenty of dislocations get tangled up in right-angle bending zone. Al4Sr particles grow in the extending zone. Finally, Al4Sr particles in products are approximately 28 μm in length. Al2Sr particles precipitate during the process. Compared with products by horizontal extrusion, Al4Sr particles by continuous extrusion are finer and distribute more evenly.

Based on the continuous extrusion machine LLJ350, experiment and Deform numerical simulation were conducted to analyze the metal flow and welding process during the continuous extrusion of AA6063 aluminium alloy with double billets. The microstructure and mechanical properties of extrusion welds were investigated through optical microscopy, scanning electron microscopy, and tensile tests. The results reveal that the oxides on the billet surface participate in the metal flow and affect the microstructure and mechanical properties of the extrusion welds. The extrusion weld exhibits bud morphological characteristics on the cross-section of the extrudate, and the welding quality of the bottom portion is superior to that of the upper portion. The welding lines are mixed with fine grains of several micrometers, and the surrounding area contains grains with a size of several hundred micrometers. When the specimens fracture at the weld, the macrofracture forms with a striped surface; by contrast, the microfracture displays a streaky structure, which includes striped protrusions and small dimples of several micron diameters. And it is also found that the extrusion welds slightly affect tensile strength, but markedly influence extrudate elongation.

Rapid and sensitive detection of the common acute lymphoblastic leukemia antigen (CD10) is crucial for diagnosing and prognosing hematopoietic tumors and several carcinomas. In the present study, a label-free immunosensor for detecting CD10 by Quartz Crystal Microbalance (QCM) was developed. The detection system has a good selectivity based on the biospecific interactions between the CD10 antigen and antibody, and the sensitivity of the immunosensor was further improved for Au NPs not only carried more antibodies but also acted as mass enhancers. Experimental results confirmed that this developed method could be conveniently used for the detection of CD10 without labeling, and CD10 could be measured quantitatively in the concentration range from 1.0 × 10 −8 to 1.0 × 10 −11 M.

Gasification of oily wastewater in supercritical water (SCW) was investigated in a reactor at 773–923 K, 25–41 MPa and residence time of 180–220 s. Pressure change had no significant influence on gasification efficiency. Higher temperature and longer residence time enhanced gasification efficiency, and lower temperature favored the production of H 2 . The effects of KOH catalyst on gas product composition were studied, and gasification efficiency was analyzed. The total organic carbon (TOC) removal, carbon gasification ratio and hydrogen gasification ratio were up to 98%, 97.88% and 128% at 923 K and 220 s, respectively, which demonstrated that oily wastewater can be completely gasified in SCW.

The effect of flow passage length in the die cavity and extrusion wheel velocity on the shape of aluminum sheath during the continuous extrusion sheathing process was analyzed by using finite element methods based on software DEFORM 3D and experimentally validated. The results show that by increasing the flow passage length, the velocity of metal at the cross-section of sheath tends toward uniformity, the values of the bending angles of sheath gradually approach the ideal value of zero and the cross-section exhibits a better shape. The extrusion wheel velocity has negligible effects on the bending shape and cross-section of the sheath product when a long flow passage is used.

Ethnopharmacological relevance Valeriana jatamansi Jones named Zhizhuxiang in China is one of the most popular traditional medicine for varied ailments related to malaise, abdominal distention, insomnia, and rheumatism. Aim of the study Evaluate the safety of iridoids rich fraction from V. jatamansi Jones (IRFV) and to provide data for clinical application. Materials and methods The acute and sub-chronic toxicity of IRFV were investigated by employing established methods. The acute toxicity study was conducted through oral administration of a single dose (3200 mg/kg body weight) of IRFV to adult mice. The vehicle used for dilution of the IRFV was a mixture of 0.5% CMC-Na and 99.5% water. The weight, diet, toxic reaction, and death after 14 days were observed. In the sub-chronic toxicity study, low doses (240 mg/kg bw), middle doses (960 mg/kg bw), and high doses (1200 mg/kg bw) of IRFV were administered daily to adult rats for 6 days a week (except Sunday) for 3 months. The general behavior of the rats was observed and recorded daily. The weight and food consumption of rats were tested weekly. The effect on organs, the hematological and blood biochemical parameters, and the histopathology were assessed after 1.5 months (five males and five females) and after 3 months (10 males and 10 females).The remaining 10 rats (five males and five females) in each group were fed for 2 weeks to observe reversible and delayed toxicity after the medicine was administered. Results In the acute toxicity study, no significant difference was found in the body weight of the mice in the control group and those in the drug group (p>0.05). The maximum tolerated dose of IRFV on mice was 3200 mg/kg, which is 2666 times of the clinical adult daily dose. In the sub-chronic toxicity study on rats, the daily single oral doses of the IRFV did not result in death nor affected the general behavior, including appearance, activities, discharge, and waste at all tested doses. Moreover, no significant difference was found (p>0.05) between the body weights of the rats from the drug groups and those from the control group. Food consumption was significantly affected (p 0.05) were observed in the hematological and blood biochemical parameters, and no abnormality of other organs were noted in both gross and histopathological examinations, except several animal transients (p

Objective: To observe the effect on TGF-beta signaling pathway of the total flavonoids from Valerian a jatamansi Jones in hepatocarcinoma 22-bearing mice. Method: Hepatocarcinoma 22-bearing mice models were established, and the mice were randomly divided into model group, tegafur group, high-dose and low-dose Valeriana jatamansi Jones group. The gene expression difference chart in TGF-beta signaling pathway in difference groups were examined by using the gene chip technology. Results: Comparing with model group, there are 7 genes expression in TGF-beta signaling pathway were significant regulated in the other three groups. The expression of Cul1, E2f5, Myc and Smad7 was up-regulated, while the expression of Comp, Smad1 and Thbs4 was down-regulated. Conclusions: The total flavonoids from Valeriana jatamansi Jones can play a important role in regulating the abnormal gene expression in TGF-beta signaling pathway in hepatocarcinoma 22-bearing mice.

An incremental target recognition algorithm based on improved discernibility matrix in rough set theory is presented. We compare the new approach with our previous nonincremental algorithm in detail, especially analyze the temporal complexities of these two algorithms. Some comparable experiments have been completed in our "Information Fusion System for Communication Interception Information (IFS/CI/sup 2/)". Experiment results show that the new algorithm is more efficient than the previous while the recognition ratio is satisfying.

An incremental target recognition algorithm based on improved discernibility matrix in rough set theory is presented. Some comparable experiments have been completed in our “Information Fusion System for Communication Interception Information (IFS/CI2)”. The results of experimentation illuminate that the new algorithm is more efficient than the previous algorithm.