Abdul Manan Ahmad

An architecture design of the intelligent agent for speech recognition and translation

An architecture design of the intelligent agent for speech recognition and translation is presented in this paper. The design involves the agent architecture and the method of the agent is used. The architecture design shows the relationship between the intelligent agent and speech recognition also translation. The intelligent agent for speech recognition is called S-AGENT and T-AGENT for translation. The purpose of the S-AGENT is to facilitate for transmitting the speech data via Internet or network. The S-AGENT is acting as a data transmit control to ensure the transmitted speech data is securely delivered. The task of the T-AGENT is different from the S-AGENT. The T-AGENT is acting as information retrieval. It processes the output from the speech recognition and translates the output based on its information memory database. If the information cannot be found on its memory, it searches the information required from the database dictionary provided. At the same time, it learns the information and saves the information to its memory for the future purpose.

Malay language text-independent speaker verification using NN-MLP classifier with MFCC

Speaker Recognition (SP) is a topic of great significance in areas of intelligent and security. In Biometric SP using automated method of verifying or recognizing the identity of the person. SP can divide into two: speaker identification and verification. In this paper we focus on speaker verification in Malay language. Speaker Verification (SV) is the task of automatically accepting or rejecting a claimed identity based on the voice characteristics of a speaker. Speaker verification can be divided into text-dependent and text-independent. In this paper, we study the applicability of Artificial Neural Network (ANNs) as core classifiers for Mel Frequency Cepstral Coefficients (MFCC). We also applied a sampled method for speaker recognition that is based on ANNs. The experiment result shows that the MLP achieved highest accuracy, the Artificial Neural Network show better performance for speech and need less training data.

User Authentication via Neural Network

The major problem in the computer system is that users are now able to access data from remote places and perform transaction online. This paper reports on the experiment and performance of using keystroke dynamics as a user authentication method. The work is designed such that it is possible for the computer system to identify authorized and unauthorized user. This is desired to control access to a system that will assign the authorized user upon entering the system. The technique used to discriminate the data is Neural Network. This paper describes the application of neural networks to the problem of identifying specific users through the typing characteristics exhibited when typing their own name. The test carried out uses two kinds of neural network model, i.e. ADALINE and Backpropagation Network. A comparison of these two techniques are presented.

Web Page Recommendation Model for Web Personalization

Web usage mining has gained more popularity among researchers in discovering the users browsing behavior mining the web server log that records all the users’ transactions activities. In this paper, we developed a usage model for predictions based on association rule. Similarity between items contained in the active user profile will be calculated upon the matched rules and finally the top-N most similar items are then recommended to the user. We used the time spent on each page for weighting the pages instead of binary. Two evaluation metrics were applied to evaluate the accuracy of the recommendations, namely precision and coverage.

An isolated speech endpoint detector using multiple speech features

Energy and zero crossing rate of the speech signal have been the two most widely used features for detecting the endpoints of an utterance. This paper proposed a new approach for locating the endpoint for isolated speech, which significantly improve the endpoint detector performance. The proposed algorithm relies on multiple speech features: root mean square energy (rmse), zero crossing rate (zcr) and cepstral coefficient (cepstrum) where the Euclidean distance measure is adopted to accurately detect the endpoint of an isolated utterance. This algorithm offers better performance than conventional algorithm which using energy only. The vocabulary for the experiment includes English digit from 1 to 9. These experimental results were conducted by 360 utterances from a male speaker. Experimental results show that the accuracy of the algorithm is quite acceptable.

Using normalize time spent on a Web page for Web personalization

The continuous growth of the information on the Internet makes it necessary for users to be provided with a convenient and yet accurate tool to capture the information needed. Web usage mining has gained more popularity among researchers in discovering the users browsing behavior. In this paper, we developed a usage model for predictions based on association rule and similarity measures. We used the normalized lime spent on each page for weighting the pages instead of binary. Two evaluation metrics is applied to evaluate the accuracy of the recommendations, namely precision and coverage. The result shows that the time spent on a Web page is essential in determining the importance of that page before it is recommended to the user.

Towards making better hybrid pattern classification design for speaker identification

Recent development on classify speaker data from a group of speaker is still insufficient to provide a satisfied result in achieving high performance pattern classification engine. There are two main difficulties in this field: how to maintain accuracy rate under incremental amounts of training data and how to reduce the time processing in the case embedded systems need to consider about efficient and simplicity of calculation. Recently we have proposed three difference hybrid pattern classification approach for text independent speaker identification system; in these approaches, we combined a hybrid GMM/VQ and decision Tree model. In this paper, we extend our investigations in order to select the most suitable hybrid GMM/VQ+DT model for real time application. For the first proposed hybrid modeling, both VQ model and GMM model will run parallel after signal preprocessing process; while the second type of proposed hybrid modeling, we present the use of decision tree in VQ techniques. The third method are extended from the second hybrid modeling which is using VQ decision rules for Gaussian mixture modeling in order to simplified the process. Experimental result shows that the third type of hybrid modeling should be considered for real world application due to the superior performance of time processing.

Adaptive Parallel Model Combination for reduced environmental mismatch in noisy speech recognition

Due to environmental mismatch, speech recognition systems often exhibit drastic performance degradation in noisy conditions. This paper presents a model based technique termed adaptive parallel model combination (APMC) which compensates the initial acoustic models to reduce the discrepancy. APMC used the well-known PMC technique to composite a set of corrupted speech models, while fine tuning the mean parameter of the models using a transformation-based adaptation technique called Maximum Likelihood Spectral Transformation (MLST). Evaluated on a context-independent phone recognition task, APMC was found to be superior to both PMC and MLST, especially in non-stationary noisy conditions. On average, APMC has achieved 48.81% improvement over the initial models, whereas PMC and MLST have improved the accuracy by 34.12% and 35.23% respectively.

Applying mobile agent in distributed speech recognition using JADE

This paper presents an application of mobile agent technology in distributed speech recognition over the internet. Mobile agent is one of a new paradigms for distributed application. With some advantages of using mobile agent for distributed application, we believe that mobile agent is a good idea to support distributed speech recognition over the internet. Speech is captured by clients, and after some local processing, the information is then sends to the server. The server recognized the speech according to an application framework and sends the result or action back to the client. In this paper, we develop the system by using JADE. JADE is one of the agents programming language and fully implemented in JAVA language.