Search

Although the security of automatic speaker verification (ASV) is seriously threatened by recently emerged adversarial attacks, there have been some countermeasures to alleviate the threat. However, many defense approaches not only require the prior knowledge of the attackers but also possess weak interpretability. To address this issue, in this paper, we propose an attacker-independent and interpretable method, named learnable mask detector (LMD), to separate adversarial examples from the genuine ones. It utilizes score variation as an indicator to detect adversarial examples, where the score variation is the absolute discrepancy between the ASV scores of an original audio recording and its transformed audio synthesized from its masked complex spectrogram. A core component of the score variation detector is to generate the masked spectrogram by a neural network. The neural network needs only genuine examples for training, which makes it an attacker-independent approach. Its interpretability lies that the neural network is trained to minimize the score variation of the targeted ASV, and maximize the number of the masked spectrogram bins of the genuine training examples. Its foundation is based on the observation that, masking out the vast majority of the spectrogram bins with little speaker information will inevitably introduce a large score variation to the adversarial example, and a small score variation to the genuine example. Experimental results with 12 attackers and two representative ASV systems show that our proposed method outperforms five state-of-the-art baselines. The extensive experimental results can also be a benchmark for the detection-based ASV defenses.

Adversarial attack approaches to speaker identification either need high computational cost or are not very effective, to our knowledge. To address this issue, in this letter, we propose a novel generation-network-based approach, called symmetric saliency-based encoder-decoder (SSED), to generate adversarial voice examples to speaker identification. It contains two novel components. First, it uses a novel saliency map decoder to learn the importance of speech samples to the decision of a targeted speaker identification system, so as to make the attacker focus on generating artificial noise to the important samples. It also proposes an angular loss function to push the speaker embedding far away from the source speaker. Our experimental results demonstrate that the proposed SSED yields the state-of-the-art performance, i.e. over 97% targeted attack success rate and a signal-to-noise level of over 39 dB on both the open-set and close-set speaker identification tasks, with a low computational cost.

The end-to-end speech translation (E2E-ST) model has gradually become a mainstream paradigm due to its low latency and less error propagation. However, it is non-trivial to train such a model well due to the task complexity and data scarcity. The speech-and-text modality differences result in the E2E-ST model performance usually inferior to the corresponding machine translation (MT) model. Based on the above observation, existing methods often use sharing mechanisms to carry out implicit knowledge transfer by imposing various constraints. However, the final model often performs worse on the MT task than the MT model trained alone, which means that the knowledge transfer ability of this method is also limited. To deal with these problems, we propose the FCCL (Fine- and Coarse- Granularity Contrastive Learning) approach for E2E-ST, which makes explicit knowledge transfer through cross-modal multi-grained contrastive learning. A key ingredient of our approach is applying contrastive learning at both sentence- and frame-level to give the comprehensive guide for extracting speech representations containing rich semantic information. In addition, we adopt a simple whitening method to alleviate the representation degeneration in the MT model, which adversely affects contrast learning. Experiments on the MuST-C benchmark show that our proposed approach significantly outperforms the state-of-the-art E2E-ST baselines on all eight language pairs. Further analysis indicates that FCCL can free up its capacity from learning grammatical structure information and force more layers to learn semantic information.

The [Nb@As8]3–anion was first isolated from solution in 1986, and a number of isostructural [M@Pn8]n−clusters (M = Nb, Cr, or Mo; Pn = As or Sb; n= 2 or 3) have since been reported. We show here how anions of this class can be used as synthetic precursors that, in combination with sources of low-valent late transition metals (Cu and Ag), generate ternary polyarsenide cluster anions with unprecedented structural motifs. Chain type [MM′As16]4–(M = Nb or Ta; M′ = Cu or Ag) units are found in compounds 2–5. These clusters contain a nortricyclane-like As7cage and a [M@As8] crown, linked by a single As atom, and represent a fusion of two quite distinct branches of polyarsenide chemistry. Our analysis of the electronic structure confirms that the cluster retains many of the features of the component units. Electrospray ionization mass spectrometry reveals a series of smaller component ions containing 8–12 As atoms, the density functional theory-computed structures of which can be understood in terms of the pseudoelement concept. This work not only presents a new type of coordination mode for As clusters but also offers a point of entry for the rational design of multinary arsenic-based materials.

Language identification (LID) involves automatically determining the language being spoken in a given segment. In code-switching speech, there are rapid switches between two or more languages within a single conversation. Despite LID attaining high accuracy on medium or long utterances, the performance on short utterances of code-switching speech is currently unsatisfactory. We propose a bidirectional encoder representation from transformers (BERT)-based LID system (BERT-LID) to enhance LID performance, especially for short-duration code-switching speech segments. The original BERT model is expanded by incorporating phonetic posterior grams (PPGs) extracted from a front-end phone recognizer as input. Then it is followed by the deployment of an optimal deep classifier for LID. Our BERT-LID model demonstrates a significant improvement of approximately 6.5% in accuracy for long-segment identification and 19.9% for short-segment identification, thereby demonstrating its effectiveness in code-switching speech LID tasks.

To discuss the application principle in tuina manipulation for lumbar intervertebral disc herniation (UDH) in Chinese literatures published in recent 30 years. The three major Chinese databases, Wanfang Academic Journal Full-text Database (Wanfang), Chongqing VIP Database (CQVIP) and China National Knowledge Infrastructure (CNKI), were searched to collect the studies of tuina manipulations in treatment of LIDH published in recent 30 years. Clustering analysis was applied to analyze the top 20 tuina manipulations for UDH. The top 20 most frequently used manipulations for LIDH were Gun-rolling, Rou-kneading, Dian-digital pressing, oblique Ban-pulling, An-pressing, Tanbo-plucking, Bashen-pulling and extending, horizontal Tui-pushing, Na-grasping, Anrou-pressing and kneading, Dou-shaking, Yao-rocking, Ca-scrubbing, Pai-patting, post-extension Ban-pulling, Mo-rubbing, Zhen-vibrating, Nie-pinching, fist-back Ji-tapping, and dorsal Shen-extending methods. The involved manipulations can be divided into two categories by the treated body areas. One category is applied to the soft tissues, including Gun-rolling, Rou-kneading, Dian-digital pressing, An-pressing, Tanbo-plucking, horizontal Tui-pushing, Na-grasping, Anrou-pressing and kneading, Ca-scrubbing, Pai-patting, Mo-rubbing, Zhen-vibrating, Nie-pinching, and fist-back Ji-tapping methods. The other category is applied to bones and joints, including oblique Ban-pulling, Bashen-pulling and extending, Dou-shaking, Yao-rocking, post-extension Ban-pulling, and dorsal Shen-extending methods. Based on the treated body area, the tuina manipulations applied to treat LIDH are predominated by the ones performed on soft tissues, assisted by those on bones and joints. From the way of force exertion, the involved manipulations are majorly the swinging methods, followed by squeezing and pressing ones. The manipulations applied to bones and joints are predominated by the Ban-pulling ones, followed by the Bashen-pulling and extending ones. 探讨近30年中文文献中推拿手法治疗腰椎间盘突出症(LIDH)的应用规律。 在万方(Wanfang)、 重庆维普(CQVIP)和中国知网(CNKI)三大中文数据库中检索近30年推拿手法治疗LIDH的文献, 采用聚类统计方法分 析治疗LIDH的前20种手法。 治疗LIDH使用频次在前20位的手法依次为法、 揉法、 点法、 斜扳法、 按法、 弹拨法、 拔伸法、 平推法、 拿法、 按揉法、 抖法、 摇法、 擦法、 拍法、 后伸扳法、 摩法、 振法、 捏法、 拳背击法和背伸法。 根据手法作用部位的不同, 可分为2类, 一类是作用于人体软组织的手法, 包括法、 揉法、 点法、 按法、弹拨法、平推法、拿法、按揉法、擦法、拍法、摩法、振法、捏法和拳背击法; 一类是作用于人体骨与关节的手法, 包括斜扳法、拔伸法、抖法、摇法、后伸扳法和背伸法。 从手法作用部位分析, 治疗LIDH的推拿手法以作用于人体软组织的手法为主, 以作用于人体骨与关节的手法为辅。从手法发力特点分析, 作用于人体软组织的手法中以摆动类手法为主, 以挤压类手法为辅; 作用于骨与关节的手法中以扳动类手法为主, 以拔伸类手法为辅。

“On water” catalytic aldol reaction catalyzed by polyetheramine (D230) has been developed for easy access to 3-substituted 3-hydroxyindolin-2-ones through the reaction between various substituted isatins and acetophenones/cyclic ketones in high yields under room temperature. Systematic mechanism investigation uncovers the secret for the on water catalytic aldol reaction: comparison of the heterogeneous and homogeneous reaction circumstances with yields of 95 and 20%, respectively, indicates the on-water reaction dominating; interfacial hydrogen bonding between isatin with H2O is tested based on the downfield shift of the C2 and C3’s 13C NMR signals when water was added to the CDCl3solution of isatin; Lewis base polyetheramine D230 catalyzes the aldol reaction via the enamine mechanism verified by in situ NMR and ESI-MS analysis.

An efficient iodine‐mediated aza‐Michael addition/C−H functionalization procedure for the synthesis of indoles was achieved in one pot. By simple aza‐Michael addition between anilines and α,β‐ynones, a series of N‐aryl enaminones intermediate was generated, followed by iodine‐mediated C−H functionalization, a wide variety of indole derivatives were obtained in moderate to excellent yields under transition‐metal‐free conditions. Control experiments and in situ ESI‐MS analysis indicated the reaction occurred via a radical mechanism. Tandem bike: A highly efficient iodine‐mediated aza‐Michael addition/C−H functionalization procedure for the synthesis of indoles from aniline and α,β‐ynones was achieved. The reaction provides access to a wide variety of indole derivatives, in moderate to excellent yields under transition‐metal‐free conditions.

Little is known about the relevance of chromogranins (Cgs) and secretogranins (Sgs) in Parkinson's disease (PD). In this study, we determined serum levels of CgA, CgB, and SgII in PD patients and assessed their association with disease severity. PD patients were recruited, identified, and classified as having early (n = 14), intermediate (n = 18), or late (n = 4) stage disease according to Hoehn‐Yahr scores. The serum concentrations of CgA, CgB, and SgII in patients with well‐defined PD (n = 36) and in healthy controls (n = 52) were measured by enzyme‐linked immunosorbent assay. Compared with controls, serum CgA levels were significantly elevated and serum SgII levels were significantly reduced in PD patients (both P< 0.05). There was no difference in serum CgB levels between the two groups. Both serum CgA and SgII levels changed progressively over time from early to intermediate to late stage (P< 0.05). Spearman correlation analysis revealed that serum CgA and SgII levels correlated with Hoehn‐Yahr and UPDRS scores (P< 0.001). These results indicate that changes in serum levels of CgA and SgII may be closely related to the severity of PD.

Articulatory features describe how articulators are involved in making sounds. Speakers often use a more exaggerated way to pronounce accented phonemes, so articulatory features can be helpful in pitch accent detection. Instead of using the actual articulatory features obtained by direct measurement of articulators, we use the posterior probabilities produced by multi-layer perceptrons (MLPs) as articulatory features. The inputs of MLPs are frame-level acoustic features pre-processed using the split temporal context-2 (STC-2) approach. The outputs are the posterior probabilities of a set of articulatory attributes. These posterior probabilities are averaged piecewise within the range of syllables and eventually act as syllable-level articulatory features. This work is the first to introduce articulatory features into pitch accent detection. Using the articulatory features extracted in this way, together with other traditional acoustic features, can improve the accuracy of pitch accent detection by about 2%.

Along with the development of the aviation industry, HM sports get more and more attention, it was a very hot sport at home and abroad, the remote control technology of model aircraft is also have been developed. In the present model aircraft remote control systems, the antenna is usually dipole antenna, dipole antenna is larger, higher power consumption . In view of the above shortcomings, this paper proposes a way that using micro-strip antenna instead of the traditional antenna. Through the discussion of the microstrip antenna's working principle, this paper designed a microstrip antenna works in 2.4 GHz model aircraft remote control system. According to its actual application , the material, shape, and the type of the microstrip antenna are determined, and the size of the antenna is calculated by the equation and the software , then the model of the antenna is built and simulated in ADS, finally the best parameters are obtained through optimizing and matching to meets the requirements.

ZigBee was rising up in recent years, It has some advantage which other wireless communication technology don’t have for example: low power consumption, short distance, networking flexible. It applies widely. This article puts forward a ZigBee household appliance control system design based on wireless communication network, through to ZigBee research. It also designed a ZigBee handheld device, which used to control domestic appliance.

The design uses ZigBee wireless networking chips to connect each street light, ZigBee coordinator of the sub-control center responsible for the formation and maintenance of street lights subnet. The ZigBee coordinator collects all streetlights information and sents to the main control center via GSM module. The main control center sends control commands to the sub-control center via GSM module, and then to each street light controller. It is low-cost, easy to install and maintain.

Atomic emission spectrometry with high accuracy, short lag time, and low detection limit, it has been widely applied in various fields. The application of atomic emission spectrometry in the converter steelmaking process has an important significance to improve the smelting of speed and steel quality. This article describes the principle of atomic emission spectroscopy (AES) analysis. Spectra obtained by the experimental for qualitative analysis, detected molten steel containing titanium element.

An experimental study of the underwater sound produced by single drop impacts influenced by different wind speeds, free fall heights and drop sizes at different depths is described in this article. Three kinds of acoustic radiation named impact sound, irregular bubble sound and delay bubble sound are distinguished by the pressure time trace. The images of bubbles taken near the point where the drop falls into water are shown. Some statistics laws of initial impact sound and irregular bubble sound affected by different condition included wind levels, heights of free fall and drop sizes are concluded. The directionality of initial impact sound is measured with wind and no wind. The process of delay bubble entrainment is analyzed and the measurements of bubble excitation probability in different condition are listed.

Using the existing GSM network for location is currently a hot, but usually the system only can complete the location function, and could not communication with others. For the purpose of the voice communication between the mobile terminals in forklifts and the service center, this paper designed a new emergency call system to overcome the above disadvantages, and especially taken their applications into consideration.

While internationally major disasters occur frequently, for any one country they are unusual events. In this project we aimed to identify public health issues arising from the physical and mental health symptoms suffered by the soldier volunteers deployed in an emergency relief task during the Wenchuan earthquake.

Energy-recycling output pad cells for driving adiabatic chips are designed, which have been fabricated with Chartered 0.35um process and tested. The proposed energy-recycling output pad cells include mainly bonding pads, electrostatic discharge (ESD) protection circuits, and two stage energy-recycling buffers that are used to drive the large load capacitances on chip pads. The two stage energy-recycling buffers are realized using CPAL (Complementary Pass-transistor Adiabatic Logic) and PAL-2N (pass-transistor adiabatic logic with NMOS pull-down configuration), respectively. For comparison, a conventional output pad cell is also embedded in the test chip. The function verifications and energy loss tests for the three output pad cells are carried out. The energy consumption of the proposed two energy-recycling output pad cells has large savings over a wide range of frequencies, as compared with the conventional CMOS counterparts, since the energy on large load capacitances in the chip pads can be well recycled.

With the decrease of the power supply voltage, the thickness of the gate oxide has been also scaled down in CMOS technologies using gate oxide materials. The leakage dissipation through the gate oxide is becoming an important component of power consumption in currently used nanometer CMOS processes without metal gate structure. Base on the fact that PMOS transistors have an order of magnitude smaller gate leakage than NMOS ones, this paper propose a P-type efficient charge recovery logic (P-ECRL) to reduce leakage dissipations in nanometer CMOS processes with gate oxide materials. For an example, a J-K flip-flop and a mode-10 counter using four-phase P-ECRL circuits are verified. All circuits are simulated using 90nm and 45nm CMOS processes with gate oxide materials. The proposed P-ECRL circuits show significant improvement in terms of power consumption over the traditional N-type ECRL counterparts.

The leakage dissipations of nano-circuits have become a critical concern. Estimating the leakage power of nano-circuits is very important in low-power design. This paper presents a new estimation technology for the active leakage dissipations of adiabatic logic circuits. Based on the power dissipation models of adiabatic circuits, active leakage dissipations are estimated by testing total leakage dissipations with additional capacitances on load nodes of the adiabatic circuits using HSPICE simulations. Taken as an example, the estimation for dynamic and active leakage power dissipations of CPAL (Complementary Pass-transistor Adiabatic Logic) circuits is demonstrated using the proposed estimation technology. The simulation results show that the proposed estimation technology can accurately estimate the active leakage dissipations of CPAL circuits with an accepted error over a wide range of frequencies.

The organometallic dimers, [{R(CO2CH2C2H-µ)2}-{Co2(CO)6}2] [R=C6H4-1,4 (1), (CH)2 (2), (CH2)2 (3)], have been prepared from reactions between Co2(CO)8 and appropriate diyne ligands in tetrahydrofuran at room temperature. These complexes have been characterized by IR spectra, 1H NMR spectra and elemental analyses. The structure of (1) was determined by single crystal X-ray diffraction. The compound (1), C26H10O16Co4, Mr=814.06, is triclinic, space group P1¯, with unit cell dimensions a=8.139(2), b=8.803(3), c=11.343(3) Å, α=99.421(6), β=96.260(6), γ=102.772(6)°. The cobalt atom is coordinated by three terminal carbonyls and an alkynyl group. The Co-Co bond length is 2.4834(7) Å.

Speech keyword search (KWS) is the task of automatically detecting the required keywords in continuous speech. Single-keyword detection can be regarded as the task of speech keyword wake-up. For many practical applications of these small vocabulary speech recognition tasks, it is costly and unnecessary to build a full large vocabulary speech recognition system. For tasks related to speech keyword search, insufficiency in data resources remains the main challenge so far. Speech pre-training has become an effective technique, showing its superiority in a variety of tasks. The key idea is to learn effective representations in settings where a large amount of unlabeled data is available to improve the performance while labeled data of downstream tasks are limited. This research focuses on the combination of unsupervised pre-training and keyword search based on the Keyword-Filler model and introduces unsupervised pre-training into speech keyword search. The research selects pre-trained model architecture Wav2vec2.0 including XLSR. The research results show that training with feature extracted by pre-trained model performs better than the baseline. In the case of low-resource condition, the baseline performance drops significantly, while the performance of the pre-trained tuned model does not decrease but even increases slightly in some intervals. It can be seen that the pre-trained model can be tuned to achieve better performance on very little data. This shows the advantage and application value of keyword search based on unsupervised pre-training.

In this paper, we propose a network for small footprint keyword spotting. It includes four parts, data augmentation, Time-Delay Neural Network (TDNN) and Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and attention mechanism. Data augmentation is Google SpecAugment with time warping and frequency mask and time mask to the spectrum on clean data set and noisy data set. TDNN and CNN model the spectrogram features from the time and space dimensions. RNN-type networks include RNN, Long Short-Term Memory network (LSTM), Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory network (BiLSTM), and Bidirectional Gated Recurrent Unit (BiGRU). The RNN extracts hidden layer features and transforms them into high-level representations. The attention mechanism is selected to generate different weights and multiplied by the high-level representation generated by the RNN to obtain a fixed-length vector. Finally, we use a linear transformation and softmax function to generate scores. We also explored the size of attention mechanism, two attention mechanisms, rectified linear unit and hidden layer of RNN. Our model has achieved a true positive rate of 99.81% at a 5% false positive rate.

A simple route to synthesise clusters [SOsCo2(CO)9] in excellent yield and mild reaction conditions is described. Several new chiral tetrahedral clusters containing the Os element were synthesised by the thermal reaction of the precursor [(μ3-S)OsCo2(CO)9] with the functionally-substituted metal exchange reagents. The structure of cluster [(μ3-S)OsCoW(CO)8C5H4C(O)CH3] was established by X-ray single crystal diffraction.

A new cluster derivative (μ3-Se)RuCoMo(CO)8CpCOOCH33was obtained by the reaction of SeRuCo2(CO)91with the functionally-substituted sodium cyclopendiene [Mo(CO)8CpCOOCH3]Na 2in THF at 60°C and its structure has been determined by single-crystal X-ray diffraction.

The new cluster [CoRuW(CO)8(μ3-S)(η5–C5H4COOCH3)] 2was isolated from the reaction of [Co2Ru(CO)9(μ3-S) 1with Na[W(CO)3(η5–C5H4COOCH3)]. Through the further reaction of cluster 2with n-propylamine, a new cluster [CoRuW(CO)8(μ3-S) (η5–C5H4CONH-C3H7)] 3was obtained and the structure of cluster 2has been established by single crystal X-ray diffraction.

The new indenyl tetrahedral cluster [(μ3-S)FeCoM (η5-Ind)(CO)8] (M=Mo 2, W 3) have been isolated from the reaction of [(μ3-S)FeCo2(CO)9] 1and K+[(η5-Ind)M(CO)3]-(M=Mo,W) under mild conditions and the crystal structure of [(μ3-S)FeCoW(η5-Ind)(CO)8] 3has been established, affording structural evidence for the ring slippage of the indenyl ligand.

The new Rh-Co mixed-metal dimer [Rh2Co2(CO)8(μ-CO)2(μ4,η2-HC≡CCH2O-)]2C6H4 4 and trimer [Rh2Co2(CO)8(μ-CO)2(μ4,η2-HC≡CCH2O-)]3B 5 have been synthesised from reactions between Rh2Co2(CO)121 and the appropriate diyne and triyne ligands. The crystal structure of cluster 4 has been determined.

The new butterfly cluster [Rh2Co2(CO)6(μ-CO)4(μ4, η2-HC≡CFeCp2)] 3 was obtained by the reaction of [Rh2Co2(CO)12] 1 with the 1-alkyne ligand HC≡CFeCp22, and the structure of cluster 3 has been established by single-crystal X-ray diffraction methods.

The new clusters [NiRuMo(CO)5(μ3-S)(η5–C5H5)(η5–C5H4R)](R = H 3 or COCH34) were isolated from the reaction of [CoRuMo(CO)8(μ3-S) (η5–C5H4R)](R = H 1 or COCH32) with NiCp2, and the structure of cluster 4 has been established by single crystal X-ray diffraction method.

ChemInform is a weekly Abstracting Service, delivering concise information at a glance that was extracted from about 200 leading journals. To access a ChemInform Abstract, please click on HTML or PDF.

Abstract Broadcast audio transcription is still a challenging problem because of the complexity of diverse speech and audio signals. Audio segmentation, which is an essential module in a broadcast audio transcription system, has benefited greatly from the development of deep learning theory. However, the need of large amounts of labeled training data becomes a bottleneck of deep learning-based audio segmentation methods. To tackle this problem, an adapted segmentation method is proposed to select speech/non-speech segments with high confidence from unlabeled training data as complements to the labeled training data. The new method relies on GMM-based speech/non-speech models trained on an utterance-by-utterance basis. The long-term information is used to choose reliable training data for speech/non-speech models from the utterances at hand. Experimental results show that this data selection method is a powerful audio segmentation algorithm of its own. We also observed that the deep neural networks trained using data selected by this method are superior to those trained with data chosen by two comparing methods. Moreover, better performance could be obtained by combining the deep learning-based audio segmentation method with the adapted data selection method. [ABSTRACT FROM AUTHOR]