Showing papers by "Mayank Dave published in 2010"

Ant colony based rule generation for reusable software component retrieval

Abstract: Storage and representation of reusable software components in software repositories to facilitate convenient identification and retrieval has been always a concern for software reuse researchers. This paper discusses and demonstrated an ant colony algorithm based technique that generates rules to store and then identify the component from software repository for possible reuse. Proposed technique help user in organization and storage of components in repository and later can help in identifying most appropriate component for given context. In first stage while searching it makes use of keywords, their synonyms and their inter-relationships. Then it makes use of ant colony optimization; initial pheromone of one is assigned to all domain representative terms of components. By updating pheromone for participating terms and non-participating terms iteratively and by calculating the quality of each rule generated, it leads to quality rules to represent and retrieve the reusable components

Fitness Evaluation of Gaussian Mixtures in Hindi Speech Recognition System

Abstract: Speech recognition (i.e. speech to text conversion)and speech synthesis (i.e. text to speech conversion)are two main operations performed in human computer interaction through natural language conversational interface. This paper presents a novel approach for modeling and designing of a Hindi speech recognition system, by using Perceptual Linear Prediction (PLPRASTA)for feature extraction and Gaussian Mixture Model (GMM) for statistical pattern classification. Experimental results show that only 4 Gaussian mixtures yield optimal performance in the context of small databases available for Indian languages which have been used to train the Hidden Markov model (HMM). The results also illustrate that the approach presented in this paper, not only outperform the traditional model for small vocabulary in typical field conditions but can also be implemented efficiently in embedded systems.

Model of Real Time Architecture for Data Placement in Wireless Sensor Networks

Abstract: Wireless sensor network (WSN) technology has promised fine grain monitoring in time and space as well as at a lower cost than is currently possible. These sensor networks are required to provide a robust service in hostile environments. Therefore the issue of real-time and reliable data delivery is extremely important for taking effective decisions in WSN. In this paper the architecture for reliable and real time approach by using sensor clusters has been proposed for storage management. Instead of storing information in an individual cluster head as suggested in some approaches, storing of information of all clusters, inside the cell is recommended within the corresponding base station. For data dissemination and action we have used Action and Relay Stations (ARS). We have developed programming model for formal specification and verification of our architecture.

CJM: A Technique to Reduce Network Traffic in P2P Systems

Abstract: In structured and unstructured Peer-to-Peer (P2P) systems frequent joining and leaving of peer nodes causes topology mismatch between the P2P logical overlay network and the physical underlay network. This topology mismatch problem generates high volume of redundant traffic in the network. This paper presents Common Junction Methodology (CJM) to optimize the overlay traffic at underlay level. CJM finds common junction between available paths and traffic is only routed through the common junction and not through the conventional identified paths. Simulation results show that CJM resolves the mismatch problem and significantly reduces redundant P2P traffic up to 87% in the best case. CJM can be implemented over structured or unstructured P2P networks and also reduces the response time by 53% approximately for the network. CJM does not alter overlay topology and performs without affecting the search scope of the network.

Distributed clustering approach for wireless sensor network based cellular data placement model

Abstract: Requiring all the nodes in a large-scale wireless sensor network to communicate their data to their respective destination will deplete the energy of the nodes quickly due to the long-distance and multi-hop nature of the communication and will also results in network contention. Therefore to increase longevity and support scalability, nodes are often grouped into disjoint and mostly non-overlapping clusters. Clustering saves energy and reduces network contention by enabling locality of communication: nodes communicate their data over shorter distances to their respective cluster-heads. The cluster-heads aggregate these data into a smaller set of meaningful information. Not all nodes, but only the cluster-heads need to communicate far distances to their respective destinations. In this paper, we propose a distributed clustering approach for our proposed real time data placement model for WSNs. It is assumed that the sensor nodes are aware of their locations in their deployment area, and they are time synchronized. For data dissemination and action in the wireless sensor network the usage of Action and Relay Stations (ARS) has been proposed.

An Empirical Approach for Optimization of Acoustic Models in Hindi Speech Recognition Systems

Abstract: A well established paradigm to develop an automatic speech recognition (ASR) system is the feature extraction at front end and liklihood evaluation of feature vectors using hidden Markov models (HMMs) with Gaussian mixtures at back end. To reduce the overall computational overhead and for proper handling of HMM parameters the appropriate selection of Gaussian mixtures and tied states is very important. This paper reviews the statistical framework and presents an empirical approach to select the optimum number of Gaussian mixtures and appropriate degree of state tying in context of small amount of training data usually available for Indian languages, specifically Hindi. At front end we have used perceptual linear prediction (PLP) combined with relative spectral (RASTA) processing for our proposed Hindi speech recognition system.

Hierarchical Data Distribution Scheme for Peer‐to‐Peer Networks

Abstract: In the past few years, peer‐to‐peer (P2P) networks have become an extremely popular mechanism for large‐scale content sharing. P2P systems have focused on specific application domains (e.g. music files, video files) or on providing file system like capabilities. P2P is a powerful paradigm, which provides a large‐scale and cost‐effective mechanism for data sharing. P2P system may be used for storing data globally. Can we implement a conventional database on P2P system? But successful implementation of conventional databases on the P2P systems is yet to be reported. In this paper we have presented the mathematical model for the replication of the partitions and presented a hierarchical based data distribution scheme for the P2P networks. We have also analyzed the resource utilization and throughput of the P2P system with respect to the availability, when a conventional database is implemented over the P2P system with variable query rate. Simulation results show that database partitions placed on the peers with higher availability factor perform better. Degradation index, throughput, resource utilization are the parameters evaluated with respect to the availability factor.

Tied Mixture Modeling in Hindi Speech Recognition System

Abstract: The goal of automatic speech recognition (ASR) is to accurately and efficiently convert a speech signal into a text message independent of the device, speaker or environment. In ASR, the speech signal is captured and parameterized at front-end and evaluated at back-end using the Gaussian mixture hidden Markov model (HMM). In statistical modeling, to handle the large number of HMM state parameters and to minimize the computation overhead, similar states are tied. In this paper we present a scheme to find the degree of mixture tying that is best suited for the small amount of training data, usually available for Indian languages. In our proposed approach, perceptual linear prediction (PLP) combined with Heteroscedastic linear discriminant analysis (HLDA) was used for feature extraction. All the experiments were conducted in general field conditions and in context of Indian languages, specifically Hindi, and for Indian speaking style.