A5102318256- Speech Recognition and Synthesis
- Music and Audio Processing
- Speech and Audio Processing
- Natural Language Processing Techniques
- Heavy metals in environment
- Chromium effects and bioremediation
- Adsorption and biosorption for pollutant removal
- Speech and dialogue systems
- Environmental remediation with nanomaterials
- Mechanical Engineering and Vibrations Research
- Radioactive element chemistry and processing
- Web Data Mining and Analysis
- Video Analysis and Summarization
- Topic Modeling
- Social Robot Interaction and HRI
- Advanced Neural Network Applications
- Vehicle Dynamics and Control Systems
- Environmental Quality and Pollution
- Water Quality Monitoring Technologies
- Soil Mechanics and Vehicle Dynamics
- Hearing Loss and Rehabilitation
- Soil and Land Suitability Analysis
- Tactile and Sensory Interactions
- Machine Learning and ELM
- Mine drainage and remediation techniques
Henan Polytechnic University
2024
Central South University
2019-2022
Northwestern Polytechnical University
2020-2022
Ministry of Education of the People's Republic of China
2019-2020
Biotechnology Research Institute
2020
Hunan Academy of Agricultural Sciences
2020
Cadmium (Cd) pollution poses a serious risk to human health and ecological security. Bioremediation can be promising effective remediation technology for treating Cd contaminated soils. In this study, seven heterotrophic strains were isolated from soil 7 autotrophic acid mine drainage. removal efficiencies compared after leaching with bacteria (Att-sys), isolates (Htt-sys) cooperative systems (Co-sys) in laboratory agitating reactors. The results indicated that efficiency of Co-sys (32.09%)...
One difficult problem of keyword spotting is how to miniaturize its memory footprint while maintain a high precision.Although convolutional neural networks have shown be effective the small-footprint problem, they still need hundreds thousands parameters achieve good performance.In this paper, we propose an efficient model based on depthwise separable convolution layers and squeeze-and-excitation blocks.Specifically, replace standard by convolution, which reduces number without significant...
Recently, several studies reported that dot-product self-attention (SA) may not be indispensable to the state-of-the-art Transformer models. Motivated by fact dense synthesizer attention (DSA), which dispenses with dot products and pairwise interactions, achieved competitive results in many language processing tasks, this paper, we first propose a DSA-based speech recognition, as an alternative SA. To reduce computational complexity improve performance, further local DSA (LDSA) restrict...
In this study soils at different depths were collected in a Zn smelting site located Zhuzhou City, China, order to understand toxic metal(loid)s distribution and microbial community vertical soil profile site. Except Soil properties content, the richness diversity of communities samples analyzed via high-throughput Illumina sequencing 16s rRNA gene amplicons. The results showed that content As, Pb, Cu, Cd, Zn, Mn was relatively high top comparison subsoil, while concentration Cr subsoil...
Abstract Voice activity detection (VAD) based on deep neural networks (DNN) have demonstrated good performance in adverse acoustic environments. Current DNN-based VAD optimizes a surrogate function, e.g., minimum cross-entropy or squared error, at given decision threshold. However, usually works on-the-fly with dynamic threshold, and the receiver operating characteristic (ROC) curve is global evaluation metric for all possible thresholds. In this paper, we propose to maximize area under ROC...
The improper stacking of chromium (Cr) slag poses a great threat to the environment and human health. toxicity Cr in soil is not only related its total amount, but also fractions. A simulated experiment was conducted laboratory assess environmental risk fractions migration distribution red soil. results showed content acid-soluble reducible significantly decreased (P < 0.05) top layer increased middle substratum layers over time. This indicated that migrated downward with time relative...
Recently, there is a research trend on ad-hoc microphone arrays. However, most was conducted simulated data. Although some data sets were collected with small number of distributed devices, they not synchronized which hinders the fundamental theoretical to To address this issue, paper presents speech corpus, named Libri-adhoc40, collects replayed Librispeech from loudspeakers by arrays 40 strongly nodes in real office environment. Besides, provide evaluation target for frontend processing...
Recently, conformer-based end-to-end automatic speech recognition, which outperforms recurrent neural network based ones, has received much attention. Although the parallel computing of conformer is more efficient than networks, computational complexity its dot-product self-attention quadratic with respect to length input feature. To reduce layer, we propose multi-head linear for reduces order. In addition, factorize feed forward module by low-rank matrix factorization, successfully number...
With the development of energy industry, power generation and transportation industry have developed from traditional petrochemical to renewable (RE). Many wind farms photovoltaic (PV) plants been set up. More more electric vehicles (EV) EV charging stations are connected grid which has brought new challenges safe stable operation grid. The grid-connected is accompanied by various faults. unstable RE also induced lots problems. In this paper an intelligent test platform was designed realize...
Self-attention (SA), which encodes vector sequences according to their pairwise similarity, is widely used in speech recognition due its strong context modeling ability. However, when applied long sequence data, accuracy reduced. This caused by the fact that weighted average operator may lead dispersion of attention distribution, results relationship between adjacent signals ignored. To address this issue, paper, we introduce relative-position-awareness self-attention (RPSA). It not only...
Large Automatic Speech Recognition (ASR) models demand a vast number of parameters, copious amounts data, and significant computational resources during the training process. However, such can merely be deployed on high-compute cloud platforms are only capable performing speech recognition tasks. This leads to high costs restricted capabilities. In this report, we initially propose elastic mixture expert (eMoE) model. model trained just once then elastically scaled in accordance with...
Self-attention (SA), which encodes vector sequences according to their pairwise similarity, is widely used in speech recognition due its strong context modeling ability. However, when applied long sequence data, accuracy reduced. This caused by the fact that weighted average operator may lead dispersion of attention distribution, results relationship between adjacent signals ignored. To address this issue, paper, we introduce relative-position-awareness self-attention (RPSA). It not only...
Transformer-based end-to-end speech recognition models have received considerable attention in recent years due to their high training speed and ability model a long-range global context. Position embedding the transformer architecture is indispensable because it provides supervision for dependency modeling between elements at different positions input sequence. To make use of time order sequence, many works inject some information about relative or absolute position element into In this...
Deep neural networks provide effective solutions to small-footprint keyword spotting (KWS). However, if training data is limited, it remains challenging achieve robust and highly accurate KWS in real-world scenarios where unseen sounds that are out of the frequently encountered. Most conventional methods aim maximize classification accuracy on set, without taking into account. To enhance robustness deep based KWS, this paper, we introduce a new loss function, named maximization area under...