Tạp chí khoa học và công nghệ

Many factors affect speech recognition including dialect factors. Vietnamese is a tone language with many different dialects. Therefore, spoken Vietnamese identification is also significantly influenced by dialect. If the information about dialects is known during the speech recognition process, the performance of recognition systems will be better because the corpus of these systems is normally organized according to different dialects. Vietnamese dialect identification has been studied and achieved certain results. Studies are usually carried out on the computer with a powerful computing environment to train models and tests. Embedded devices are being used extensively in practical products, including speech recognition and control, which often have weaker configurations than computers such as slow speed, low memory. Therefore, the implementation of speech recognition on this hardware is more a little difficult. This article presents the results of the research and experiment of Vietnamese dialect identification at embedded systems using the VDSPEC corpus on Raspberry embedded KIT. The study used the GMM identification model with the parameter set of 13 MFCC coefficients and the F0 parameter was standardized according to the average F0 of each sentence, the average recognition rate of dialects is 70%.

We have stupied structure and optical properties of ZnO - SiO2 nanocomposites doped with Eu3+ ions with varied concentrations of ZnO and the annealing temperatures. The thin films containing ZnO - SiO2 nanocomposites doped Eu3+ ions were preparedby sol – gel method and spin-coating process. The stuctures and morphologies of the thin films are studied by X- ray diffraction ( XDR) and scanning electron microsope (SEM). SEM images and X- ray diffraction patterns indicate that ZnO nanocrystals were creatin SiO2 matrix with particle size of about 3- 5 nm. In addition, the emission and excitation spectra were studies. From the analysis of the optical data, we are able to demontrate that multiple Eu- related optical centers are present in the studied samples. The dominating of 5D0 – 7F2 transition of Eu3+ ions at about 613 nm suggests that Eu3+ ions mainly take a site with inversion antisymmetry in the ZnO host. Also, the energy transfer from ZnO nanocsystals ta Eu3+ ions are observed in the excitation spectra with the excitation wavelengths ranging from 250 nm to 450 nm.

One of the major challenges of Mobile Ad hoc Network (MANETs) is how to implement and avoid of different kind of security attack such as Denial of service (DoS), wormhole, replay, masquerade, black hole etc. MANETs use routing protocol to communicate the data packet from one node to another and each node acts as a host or router, which can choose suitable paths for transmission of packet. AODV is one of the most popular used routing protocols in ad-hoc networks, but it also has a lot of potential lack of security. The goal of attacks is usually to disrupt the operation of the network or to affect the network performance. In this paper, we implement and analyses performance of multiple black hole attack nodes simultaneously in AODV routing protocol in terms of throughput, energy efficiency and data packet delivery. Our simulation results show that the more attackable nodes in network simultaneously, the lower the energy efficiency and network performance

In an Internet of Thing (IoT) environment, entities with different attributes and capacities are going to be connected in a highly connected fashion. Specifically, not only the mechanical and electronic devices but also other entities such as people, locations, and applications are connected to each other. Most IoT applications must work with dynamic and speedily changing systems due to new entities are coming online and/or the connection between these entities can change regularly. This requires a data model that enables to easily represent the entities and support adding, deleting, and updating relations between entities without impacting application availability. Fortunately, graph databases are purposely-built to store highly connected data with nodes representing entities and edges representing relationships between these entities. In this paper, we propose a general graph model that can be used to design graph databases in order to support effectively storing and analyzing IoT data. We represent IoT data based on a graph model and consider smart building data management as a case study. Through the analysis and comparison of experimental results in various aspects, we find that our graph modeling approach is applicable for storing and analyzing the IoT connected data.

This article presents the studying results of surveying 301 female diabetics about the request of their shoes into: shoe type, shoe kind and shape, the dimension and shapes of shoe’s details, the bottoming assembly method, the materials and the comfort of the shoes. The study results show that: popular shoes for female diabetics are open shoes, low-necked with Velcro, low heel from 20mm to 25mm, high toe spring from 10mm to 15 mm, and large, high and oval toe part. Materials usually chosen are: The shoe uppers part mainly from leather, shoe upper lining and insole from antibacterial fabric, sole from porous materials such as foam rubber, EVA foam, PU foam. Most diabetic women surveyed selected following criteria: soft shoes, elastic, fit feet to create comfort while moving. The study results are the base to build the requirement about structure and materials for designing and making special shoes for female diabetics in Vietnam.

Mạng nơ ron học sâu được ứng dụng trong nhận diện hình ảnh, nhận diện giọng nói, xử lý ngôn ngữ tự nhiên. Hiện nay rất nhiều các bài toán nhận dạng sử dụng mạng nơ ron học sâu, vì nó có thể giải quyết các bài toán với số lượng lớn các biến, tham số kích thước đầu vào lớn với hiệu năng cũng như độ chính xác vượt trội so với các phương pháp phân lớp truyền thống, xây dựng những hệ thống thông minh với độ chính xác cao. Trong bài báo này, các tác giả nghiên cứu mạng nơ ron học sâu (CNN - Convolutional Neural Network)[1] cho bài toán phát hiện lửa cũng như khói bằng việc xử lí dữ liệu từ camera giám sát nhằm phát hiện, cảnh báo cháy rừng.

This work showed that, the NiO / PPy hybrid films were successfully synthesized by a simple direction from vapor phase polymerization at room temperature and normal pressure. The morphological and structural characteristics of these hybrid films have been examined and compared with the pure PPy and NiO films through SEM image, X-ray diffraction and Fourier-transform infrared spectroscope (FTIR). The measurement results of the initial NH3 gas sensitivity property indicated the response of the NiO/PPy hybrid film is remarkably higher than that of the PPy film without hybridization at room temperture. This research has shown the applied potential in gas sensor at room temperature based on the hybrid material groups between conductive polymers and transition metal oxides.

In this article, we investigate a risk model with a quota-(α, β) reinsurance contract. The premium process and claim process are assumed to be independent sequences of indentically distributed random variables. Using inductive method, we obtain upper bound of infinite-time ruin probability of an insurance company.

In this paper, a combination between the indentation and scratching process was developed to analyze the deformation mechanisms and mechanical properties of Cu50Zr50 metallic glasses (MGs) using molecular dynamics (MD) simulation. The deformation mechanisms and mechanical properties of Cu50Zr50 MGs are appraised through the surface morphology, pile-up height, hardness, machining forces, and resistance coefficient. The influences of different indenter radius are clearly investigated. The results exhibit that the machining zone increases as increasing indenter radius. The pile-up height and hardness reduce with a bigger radius of the indenter. The hardness values ​​range from 7.94 to 13.33 GPa. The forces increase, however, the resistance coefficient decreases as the indenter radius increases.

Bài báo này giới thiệu một phương pháp tóm tắt trích rút các văn bản sử dụng BERT. Để làm điều này, các tác giả biểu diễn bài toán tóm tắt trích rút dưới dạng phân lớp nhị phân mức câu. Các câu sẽ được biểu diễn dưới dạng vector đặc trưng sử dụng BERT, sau đó được phân lớp để chọn ra những câu quan trọng làm bản tóm tắt. Chúng tôi thử nghiệm phương pháp trên 3 tập dữ liệu với 2 ngôn ngữ (Tiếng Anh và Tiếng Việt). Kết quả thực nghiệm cho thấy phương pháp cho kết quả tốt so với các mô hình khác.