Conventional automatic speech recognition (ASR) systems often encounter difficulties in processing speech of dysarthric people, which is characterized by impaired neuromuscular control and atypical articulation. Existing commercial solutions, such as Google Speech to Text and Microsoft Azure Speech, exhibit high error rates when processing the speech of people with dysarthria, severely hindering the accessibility of communication. This study introduces a novel personalized ASR system that aims to address these challenges by integrating speaker adaptive modeling and advanced correction mechanisms. The system is based on the transformer-based SenseVoice mini-model, which utilizes the functionality of the transformer-based Large Language Model (LLM) to extract speaker-specific speech features from historical data and implement a dynamic history correction algorithm. By employing an LLM-based style extraction method, the system develops a personalized correction model that can be adapted to individual speech variations. Experimental evaluations show that the system provides substantial improvements in recognition accuracy and processing efficiency compared to traditional automatic speech recognition (ASR) techniques. This research highlights the potential of AI-driven personalization approaches in assistive communication technologies, providing a promising avenue for enhancing communication tools for people with speech impairments.

Due to the intensification of global warming, changes in sea temperature and salinity have profound impacts on ecosystems and climate. Accurately predicting marine environmental changes has thus become an important issue. However, the spatial limitations inherent in Argo floats and satellite remote sensing data pose challenges to prediction. This study integrates data collected from Argo floats and satellite remote sensing, employing a Transformer-based model to predict sea temperature and salinity at different times, latitudes, and depths below 50 meters beneath the sea surface, aiming to supplement the shortcomings of Argo floats data and satellite remote sensing data and achieve comprehensive marine environmental prediction. The results of the study demonstrate that the proposed Transformer-based model performs well in sea temperature prediction, with a mean absolute error (MAE) of 0.489°C, root mean square error (RMSE) of 0.809°C, and the coefficient of determination (R2) of 0.983. Regarding salinity prediction, the model also exhibits excellent performance, with MAE of 0.167 psu, RMSE of 0.229 psu, and R2 of 0.805. Compared to other models, the model proposed in this study shows superior performance in predicting sea temperature and salinity, providing an important tool and reference for future marine resource management and climate change research.

Air pollution has witnessed that is an important issue regarding with people health. An excellent air pollution forecasting approach can provide warning to user for avoiding too much polluted air inhalation. Many studies have proposed the model of predicting PM2.5 concentration. How to improve the performance of the forecasting models is an important issue. As well known, ensemble learning is proved that it can improve the forecasting performance. In this paper, in order to improve the performance of air pollution forecasting, we explore an approach named hybrid model framework. which exploit the stacking scheme of ensemble learning. In this work, first we proposed the framework used to forecast the air pollution based on traditional machine approach and deep learning; and exploited Pearson correlation coefficient to calculate the correlation of various models, and finally find out the highest correlation among these models. We conducted experiments and made a comparison with four kinds of models to demonstrate the forecasting performance of hybrid model. The experimental results reveal that the hybrid model is superior than single air pollution forecasting model.

As the advance of technology, it is increasingly speech recognition tools on mobile devices, such as Google Voice and Apple Siri, those have been widely used and have high recognition rate of people's speech. However, these speech recognition tools cannot work well for people with the disease of "Dysphasia" and has very low recognition rate. It is important issue to develop a speech recognition toot for dysphasia, such as Cerebral Palsy(CP) and Amyotrophic Lateral Sclerosis(ALS), to assist those people communicating with others well. Recently, there are various open source programs has been announced, such as Google's Tensorflow, which is used to develop speech recognition based on Deep Neural Networks (DNNs). In this paper, we propose a Convolutional Neural Networks (CNNs) model to perform speech recognition for ALS. The model consists of two hidden CNN layers, each includes one CNN, Rectified Linear Unit (ReLU), Dropout Unit, and MaxPooling Layer. We implement the CNN for dysphasia speech recognition using Google's Tensorflow and collect 33 pronunciations from a dysphasia, each pronunciation at least has 350 training voice files. The highest accuracy is about 63% for one word. And it will be up to totally 94% for top five words. The result shows that it can effectively recognize speech for dysphasia. It can be expected to construct a speech recognition system for assisting dysphasia communicating with others.

It is shown that the excessive inhalation of PM (Particulate Matter) 2.5 will seriously affect the health of human. Many countries have deployed various detectors for air pollution in order to report concentration of PM2.5 to show how much seriousness of air pollution is. But, what more important is how much PM 2.5 has been inhaled by people anytime and anywhere. Therefore, in this paper, we propose a method for monitoring and estimating people's PM2.5 inhalation by using personally physiological data, such as heart rate (HR), and mobile PM2.5 sensor. We can retrieve people's heart rate from company's cloud platform of smart bracelet and then calculate his/her PM2.5 inhalation based on personal HR and minute ventilation (VE). In the method, we first adopt Bruce protocol to measure various functions of body, such as HR and VE, on a treadmill and exploit regression model to build predictive model for VE. Therefore, we can utilize smart bracelet to monitor people heart rate and then calculate PM2.5 inhalation based on personal HR and mobile PM2.5 sensor (MPS). We think that the method can effectively estimate the PM 2.5 inhalation of physical activity.

Online posts not only represent the records of people's lives but also reveal their satisfaction with life and relationships as well as potential mental illnesses. The detection of (strong or general) negative as well as (strong or general) positive feelings of people from online posts can keep us from carelessly missing their important moments, difficult or great, due to the overloaded information in the daily life and lead to a better society. Therefore, in this paper, we build a Feeling Distinguisher system based on supervised Latent Dirichlet Allocation (sLDA), Latent Dirichlet Allocation, and SentiWordNet methodologies for detecting a person's intention and intensity of feelings through the analysis of his/her online posts. Experimental results on posts collected from five social network websites demonstrate the effectiveness of FeD. The performance of FeD is about 1.08-1.18 folds that of SVM and sLDA.

Due to the emergence of social platforms, people tend to posting their diaries and feeling online for sharing with others. In this paper, we aim to predict whether a user is getting depressed or not through his blog posts on the Internet. For this purpose, we use Latent Dirichlet Allocation (LDA) to find out top frequency words appearing in a user's diaries and use SentiWordNet to calculate the emotion score of the user. Experimental results show that our method is useful in the diagnosis of mental disorder detection in social platforms.

With the advances of information technology, there are increasing researches aiming at assisting depression diagnosis and treatment. In most of them the user is necessarily actively joining the diagnosis and treatment program while he has perceived mental disorder himself. In order to early prevent the mental disorder, in this paper we propose an early warning mechanism that observes and mines user diary published on social network platform, and generates a score of getting mental disordered or depressed. If the score is large than a threshold, the system can notify the user and his friends on the social network to take care about the friend. We have conducted experiments to evaluate the proposed approach, and the results show that the proposed approach is effective.

Recently, depression become a general disease in the world due to the promotion of life quality and technology development. Most of people are not aware of the possibility of getting depressed himself in daily life. To accurately diagnose getting depressed becomes an important issue. In this paper, we utilize ontologies and Bayesian networks techniques to build the inference model for inferring the possibility of depression. We propose an ontology model to build the terminology of depression and utilize the Bayesian networks to infer the probability of depression. In addition, the paper also proposes an agent-based platform and addresses the implementation issue. The result shows that it can be well-inferring in the depression diagnosis.

With the advance of cloud computing, cloud service providers (CSP) provide increasingly diversified services to users. To utilize general search engine, such as Google, is not an effective and efficient if the services are similar but with different attributes. Therefore, an intelligent service discovery platform is necessary for seeking suitable services accurately and quickly. In the paper1, we propose a framework that integrates intelligent agent and ontology for service discovery in cloud environment. The framework contains some agents and mainly assists users to discovering suitable service according to user demand. User can submit his flat-text based request for discovering required service. We implement a cloud service discovery environment to demonstrate the concept and its application. We also utilize the Recall and Precision to evaluate the accuracy of the system.

With the advance of cloud computing, hybrid cloud that integrate private and public cloud is increasingly becoming an important research issue. Migrating cloud applications from a busy host to an idle host needs an efficient way to guarantee the performance in the geographical heterogeneous cloud environment. This paper we propose an automatic, intelligent service migration framework on a hybrid cloud based on agent technology. We build a prototype that integrated our private cloud with public cloud. In the prototype, mobile agent technique is exploited to manage all resources, monitor system behaviour, and negotiate all actions in the hybrid cloud, in order to achieve automatic, intelligent service migration between the clouds. We demonstrate the service migration mechanism on Hadoop platform between our platform and ITRI's public cloud.

It is increasingly concerning global change and its regional impacts. There are many scientific information systems focused on this issue. In this paper1, we, based on the metadata classification and ontology, propose a scientific information retrieving and extracting architecture that can tie the metadata classification mechanism to extract the searched file efficiently and can accurately access and view Argo' data. The architecture is built by utilizing mediator/wrapper architecture to develop a scientific data extracting system. The system can easily help oceanographer to analyze the ocean's ecology by means of temperature, salinity and other information. The result of performance evaluation shows that one is about the architecture with the help of metadata classification can extract user's desired data effectively and efficiently, and the other is concerned about correctness of information retrieval.

In the paper1, we propose a rule-based mobile agent generation framework for information retrieval in a flexible, transparent, and easy way. User can submit a flat-text based request, the request will be automatically deduced by a Reasoning Agent (RA) based on predefined ontology and inference rule, and then be translated to a Mobile Information Retrieving Agent Description File (MIRADF) that is formatted in a proposed Mobile Agent Description Language (MADF). A generating agent, named MIRA-GA, is also implemented to generate a MIRA according to the MIRADF. We also design and implement a prototype to integrate these agents and show an interesting example to demonstrate the feasibility of the architecture.

Near Field Communication (NFC) is a two-way communication technology based on radio frequency identification (RFID), which in recent years has become one of the most popular devices in a great variety of applications. The NFC-equipped phone (NFC phone) that embeds NFC technology into a cellular phone makes it increasing attractive for business uses, such as for making payments or ticketing. This paper presents a novel architecture for NFC phone-driven, personalized, context-aware smart spaces. With this architecture, users can employ the phone, which carries predefined personal desires, to control devices, such as home appliances. The appliances are automatically controlled and driven in response to a request sent from the phone by using predefined ontology and rule-based reasoning. We also implement a prototype to demonstrate the feasibility of the architecture, evaluate its performance, and show its efficiency.

Artificial neural networks (ANN) are developed for highly parallel and distributed systems. These systems are able to learn from experience and to perform inferences. Although Petri nets (PNs) were modified to be ANN-like multilayered architectures for fuzzy reasoning, some researchers have paid more attention to the PN-based learning so far. In this paper, we have developed supervised and unsupervised learning algorithms for the machine learning PN (MLPN) models in order to make them fully trainable and to remedy the difficulties encountered by ANN. When compared with ANN, the MLPN model shows some significant advantages. Main results are presented in the form of five observations and supported by some experiments.

In this paper, we, based on mobile agent technology, propose a resource-aware knowledge extraction architecture on mobile grid, named RAKER. RAKER can dynamically determine the processing and policy for achieving high-performance and high availability of knowledge extracting based on our previous proposed resource estimation model. On the RAKER, users can extract information or knowledge in efficient,effective and transparent way that kept on the mobile grid without caring about the energy consumption that is most important issue in mobile computing. We show the implementation and an example to demonstrate the use of RAKER. In addition, we also measure the latency and the energy consumption and simulate the system availability to show that the performance of RAKER.

Argo project is an international ocean-observatory project that has a global array of 3,000 more free-drifting profiling floats. Argo data is a large collection of data files. To retrieve Argo data from the large database is a complicated work even if the Argo database has a metadata file per day for helping user to inquire the target files. This paper proposes a metadata miner (MM) approach to assist user program to inquire the target files quickly. In addition, we also propose an integrated information retrieval framework that based on mediator/wrapper approach to smoothly tie the MM. According to the performance evaluation, it shows that the MM approach has a significant performance enhancement in finding the target file.

In this paper, we propose an agent and ontology based information retrieval architecture for mobile grid, named AIR, to discover and extract from mobile grid environment in a flexible, transparent, and easy way. Based on the AIR, users can submit a flat-text based request to mobile grid for discovering information or knowledge. The request will be automatically reasoned by a reasoning agent based on ontology and reasoning rule, and be translated to a mobile information retrieving agent (MIRA) that can be migrated to mobile grid node to retrieve information that users require. In order to generate a mobile IR agent, we propose a mobile agent description language (MADL) that defines the behavior of mobile IR agent. In addition, a knowledge synthesis agent in the architecture is designed to collect and synthesize information from mobile grids. Finally, we show an interesting example to demonstrate the feasibility of the architecture.

The distributed object oriented computing model is the next logical step to develop distributed applications. In recent years, several object models have been proposed, such as COM/DCOM, CORBA, and JAVA Bean etc. In CORBA, which was announced by OMG, object request broker is a software bus to connect applications and object components. In addition, multi threaded programming is a well known technique to improve the performance of applications. In a CORBA environment, clients can invoke the remote objects that are shared. If those objects are single threaded it will affect system performance in large distributed applications. We describe in detail the design and implementation of multi threaded object request broker based on CORBA. Our ORB was implemented atop Windows NT and underlying TCP protocol. Finally, we compare our system's performance with IONA's Orbix, which is a well known commercial product, in both one-way and two-way request.

The concept of transactions is not only indispensable in database applications, but also useful in building robust software for mission critical applications. The paper presents an implementation of the Object Transaction Service (OTS) based on CORBA 2.0 specification. Transactional applications developed with the support of our OTS implementation are able to assure the ACID properties even in the presence of node crashes, software system failures and process hangs. The preliminary results obtained from the experiments on Sun workstations with Orbix 1.3 show that the overhead due to the OTS service is satisfactory for most applications.

There are more and more COSS (Common Object Service Specifications) on CORBA (Common Object Request Broker Architecture) announced by the OMG (Object Management Group), but no common specification about fault-tolerance exists. We propose a "warm stand-by" replication approach. When an object (primary object) is invoked, it will invoke a secondary object, and the primary object will log the messages and checkpoint the state to the secondary object periodically. If the primary object fails, the secondary object can take over by way of a client executing a few operations to change the secondary object's mode to primary. Following the style of COSS, we define four interfaces and provide class implementations that can help programmers write programs with fault-tolerant capability. The whole model has been implemented on Orbix, which is a full implementation of CORBA specification.