IPRC - 2015
Permanent URI for this collectionhttp://repository.kln.ac.lk/handle/123456789/156
Browse
4 results
Search Results
Item Hardware Implementation of a Hidden Markov Model Based, Speaker Independent, Continuous, Sinhala Speech Recognition System(Faculty of Graduate Studies, University of Kelaniya, 2015) Samankula, W.G.D.M.; Dias, N.G.J.A speaker independent speech recognition system is built to recognize the continuous Sinhala speech sentences using the toolkit, HTK 3.4.1 based on the statistical approach, Hidden Markov Model (HMM). Mel Frequency Cepstral Coefficient (MFCC), Perceptual Linear Prediction (PLP) and Linear Predictive Coding (LPC) are considered as the feature extraction methods. The recognition performance is considered for number of feature parameters varied from 4 to 12, by adding energy coefficients, first and second derivatives of each coefficient, in order to find the optimal number of parameters for each feature extraction method. Three hundred Sinhala sentences were considered for recording in order to create the phonetically balanced dictionary. Data recordings were done with 50 males and 50 females and testing was performed by 25 speakers who had participated and had not participated for the training. The recognized sequence of words are the commands to automate home appliances such as light, television and radio etc., and this can help people with motor disabilities to operate equipment. The speech recognition system was physically implemented to provide access from a PC or a laptop, based on Arduino UNO board (ATmega328 microcontroller). Arduino comes with a simple integrated development environment (IDE) and allows the programmer to write programs for Arduino in C language. The identified command is transferred to the Arduino UNO board through serial communication and the signal is transmitted using Radio Frequency (RF) to operate electrical home appliances from anywhere up to 150 meters using wireless transceiver modules (C1101) with operating frequency 433MHz. Software was developed to operate more than 18 home appliances, but in hardware implementation, only four are tested. Four Arduino UNO boards are used to implement the light and fan on/off control and the door and curtain angle control. On/off control is operated using relays to switch on and switch off. The door and curtain angle control are constructed by motor with the MOSFET transistors (IRFZ44N). Since a high recognition rate of 85% was achieved for MFCC with 7 feature parameters and adding energy coefficients, first and second derivatives in the software analysis of the previous studies, the same model was used to implement the hardware. A different grammar file is created in the language model of the software to achieve high recognition rate, by considering words and phrases that are only needed to operate the hardware.Item Deep Unsupervised Pre-trained Neural Network for Human Gesture Recognition(Faculty of Graduate Studies, University of Kelaniya, 2015) Kumarika, B.M.T.; Dias, N.G.J.Recognition of visual patterns for real world applications is a complex process that involves many issues. Varying and complex backgrounds, bad lighting environments, person independent gesture recognition and the computational costs are some of the issues in this process. Since human gestures are perceived through vision, it is a subject of visual pattern recognition. Hand gesture recognition is of higher interest for Human-Computer Interaction (HCI), due to its widespread applications in virtual reality, sign language recognition, robot control, medical industry and computer games. The main goal of the research is to propose a computationally efficient and accurate pattern recognition algorithm for HCI. Deep learning attempts to model high-level abstractions (features) in data and build strong feature space for the recognition task. Neural network with five hidden layers was used and each layer can learn features at a different level of abstraction. However, training neural networks with multiple hidden layers was difficult in practice. At first, each hidden layer individually was trained in an unsupervised fashion using autoencoders. After training the first autoencoder, second autoencoder was trained in a similar way. The main difference is that features that were generated from the first autoencoder are used as the training data in the second autoencoder thus decreased the size of the hidden representation, so that the second autoencoder learns an even smaller representation of the input data. The original vectors in the training data had 101376 dimensions. After passing them through the first encoder, this was reduced to 10000 dimensions. After using the second encoder, this was reduced to 1000 dimensions. Likewise at the end, final layer was trained to classify 50 dimensional vectors into different image classes. The result for the deep neural network is improved by performing Backpropagation on the whole multilayer network. Finally, we observed that average test classification error for traditional neural network with supervised learning algorithm is 3.6% while the error for pre-trained deep neural network is 1.4%. We can conclude that unsupervised pre-training adds robustness to a deep architecture and it proposes computationally efficient and accurate pattern recognition algorithms for HCI.Item Application of written-bell discounting techniques for smoothing in part of speech tagging algorithm for Sinhala language(Faculty of Graduate Studies, University of Kelaniya, 2015) Jayaweera, M.P.; Dias, N.G.J.Item A Plug-in to Boost the Behaviour of a Rule-Based Expert System More Like a Human(Faculty of Graduate Studies, University of Kelaniya, 2015) Weerakoon, W.A.C.; Karunananda, A.S.; Dias, N.G.J.Artificial Intelligence (AI) is one major aspect of Computer Science. Among the applications of AI, expert systems are predominant. There are expert systems built for variety of subject domains such as education, medicine, and engineering, and were built by imitating the human experts with the ability to make accurate decisions by resolving the proper set of rules and facts stored in a knowledgebase to solve more complex problems. When it comes to systems, it is expected to be more accurate, reliable, efficient and complete. The current expert systems consists of many facilities such as user interfaces, reasoning of the system, knowledgebase, working memory, making inferences, prioritizing and an automatic way for the user to enter knowledge, with compared to the human experts. Even though, the expert systems are still behind and much specific in some aspect such as the abilities in generalizing concepts, drawing associations among knowledge entities depending on the causal relationships, adding new knowledge, removing irrelevant knowledge, prioritizing knowledge entities for the execution as per the input to gain improvements over generations of execution as human experts do. Among the technical categories of the expert systems such as rulebased, frame-based and induction-based, our concern is to improve the rule-based expert systems by solving the said problem by constructing a processing model which consists of the processing states such as Origin, Classified, Pre-State, Resolve and Terminate with newly introduced multiple sub-processes such as Input/Identify knowledge entities, Classify facts/rules depending on the causal relationships crafting the generalizing facility and Termination. When the system executes over generations, it produces outputs and gains improvements using the above mentioned processing model as per the input/queries. For this processing model, newly introduced sub-processes will be implemented using C programming language and will integrate to the current expert systems written in ‗C Language Integrated Production System‘ as a plug-in. The system will be able to evaluate by comparing its states With-Plug-In and Without-Plug-In for the quality using a non-parametric test such as Mann-Whitney-U-test and for the time using a paired-t-test. As a result we are capable of providing an expert system which is more like a human expert.