Junaid Akhtar

A novel approach to increase the robustness of speaker independent Arabic speech recognition

This work presents a two-tier approach through sequential application of intensity contours and formant tracks for accurate Arabic phoneme identification. The recognition system developed is based on data sets of 40 speakers for each Arabic phonetic sound. As a first step towards recognition of phonemes, the sound is sampled and then preprocessed to get formant frequencies and intensity contours. In order to automate the intensity and formant based feature extraction, a generalized regression neural network has been implemented, trained and validated on 21 input features.

A framework for evolutionary algorithms based on Charles Sanders Peirce's evolutionary semiotics

One of the objectives of Evolutionary Computation (EC) has been to understand the processes of natural evolution and then model them algorithmically. Hans-Paul Schwefel, in his 1997 paper on the future challenges for EC argues that the more an algorithm models natural evolution at work in the universe, the better it will perform (even in terms of function optimization). There is enough data to suggest that slight differences in the understanding of the natural evolution can cause the associated Evolutionary Algorithms (EA) to change characteristically. The present paper tests Schwefels hypothesis against Charles Sanders Peirces theory which places semiotics, the theory of signs, at the heart of universal evolution. This course is followed because of three primary reasons. Firstly, Peirce has not been seriously tested in EC, although there have been EA based on other theories and sub-theories. Secondly, Peirces universal theory, by not being restricted to biological evolution alone, qualifies for Schwefels hypothesis, perhaps more than most other theories that have already been modeled algorithmically. But most importantly because, in experimental terms, it warrants an original claim that Peirces insights are useful in improving the existing EA in computer science, as Peircean EA can potentially solve some of the major problems in this area such as the loss of diversity, stagnation, or premature convergence. This paper provides a novel framework and consequently a simple algorithm based on Peirces theory of evolution, and tests it extensively against a benchmark set of mathematical problems of varying dimensions and complexity. Comparative results with classical and advanced EA form another significant part of the paper, and help in strengthening the viability of Schwefel-Peirce hypothesis for EC.

Content Based Video Retrieval Using Particle Swarm Optimization

Traditional video search engines retrieve the results on the basis of correspondence between users textual query and tags associated with the videos. Only that content that matches the tags is returned as a result to the user. Given the ever-increasing immensity of videos on the internet, especially those with zero or irrelevant tags, such traditional methodology has eventually led to rise in ratio of missing important context. Content based searching within a video library is definitely an alternative solution but it requires time consuming computations and comparisons which renders exhaustive search unpractical. The purpose of this paper is to provide an efficient methodology that will lead to incremental improvement in the video search results against a users query image. Our method employs Particle Swarm Optimization (PSO), an evolutionary population based search algorithm, to look for frames within the video library. The fitness of each swarm particle is the degree of similarity with respect to the content present in both the input image provided by the user and the video frame(s) fetched through PSO. This exempts us from the exhaustive and linear search of every frame of every video in the library. The relative best match in each generation of PSO is shown to the user for his engagement. For calculating the fitness of each swarm particle we have tested three similarity measures, 1) correlation based template matching, 2) score from scale-invariant feature transform (SIFT) algorithm and, 3) convolution. Preliminary results on real video library are promising.

Sign language localization: Learning to eliminate language dialects

Machine translation of sign language into spoken languages is an important yet non-trivial task. The sheer variety of dialects that exist in any sign language makes it only harder to come up with a generalized sign language classification system. Though a lot of work has been done in this area previously but most of the approaches rely on intrusive hardware in the form of wired or colored gloves or are specific language/dialect dependent for accurate sign language interpretation. We propose a cost-effective, non-intrusive webcam based solution in which a person from any part of the world can train our system to make it learn the sign language in their own specific dialect, so that our software can then correctly translate the hand signs into a commonly spoken language, such as English. Image based hand gesture recognition carries sheer importance in this task. The heart of hand gesture recognition systems is the detection and extraction of the sign (hand gesture) from the input image stream. Our work uses functions like skin color based thresholding, contour detection and convexity defect for detection of hands and identification of important points on the hand respectively. The distance of these important contour points from the centroid of the hand becomes our feature vector against which we train our neural network. The system works in two phases. In the training phase the correspondence between users hand gestures against each sign language symbol is learnt using a feed forward neural network with back propagation learning algorithm. Once the training is complete, user is free to use our system for translation or communication with other people. Experimental results based on training and testing the system with numerous users show that the proposed method can work well for dialect-free sign language translation (numerals and alphabets) and gives us average recognition accuracies of around 65% and 55% with the maximum recognition accuracies rising upto 77% and 62% respectively.

Evolutionary Algorithms based on non-Darwinian theories of evolution

One name that comes to mind in connection with the word evolution is Darwin. One evolutionist however, who is rarely talked about, especially in the Artificial Intelligence community, is Peirce. The Darwinian model is based on the concepts of absolute chance, mechanistic laws, and inexplicable interaction between the two. In contrast, Peircepsilas framework posits a dynamic interaction between possibility, necessity and regularity to describe the process of evolution. The theory of evolution proposed by Peirce is superior to the one proposed by Darwin because it is more general and it has greater explanatory power. Peircepsilas insights are significant enough to be used to improve the existing evolutionary algorithms. It was observed during our literature review that almost all evolutionary algorithms are fundamentally based on Darwinian principles of evolution. The present paper highlights the differences between Darwinian and Peircian evolutionary theories and provides the theoretical foundation for developing a novel Peirce based Evolutionary Algorithm. Preliminary experiments have been conducted and results seem very promising.

A Hybrid Multi-layered Speaker Independent Arabic Phoneme Identification System

A phoneme identification system for Arabic language has been developed. It is based on a hybrid approach that incorporates two levels of phoneme identification. In the first layer power spectral information, efficiently condensed through the use of singular value decomposition, is utilized to train separate self-organizing maps for identifying each Arabic phoneme. This is followed by a second layer of identification, based on similarity metric, that compares the standard pitch contours of phonemes with the pitch contours of the input sound. The second layer performs the identification in case the first layer generates multiple classifications of the same input sound. The system has been developed using utterances of twenty-eight Arabic phonemes from over a hundred speakers. The identification accuracy based on the first layer alone was recorded at 71%, which increased to 91% with the addition of the second identification layer. The introduction of singular values for training instead of power spectral densities directly has resulted in reduction of training and recognition times for self-organizing maps by 80% and 89% respectively. The research concludes that power spectral densities along with the pitch information result in an acceptable and robust identification system for the Arabic phonemes.

An evolutionary algorithm derived from Charles Sanders Peirce's theory of universal evolution

Historically, Evolutionary Algorithms (EAs) have been important for Evolutionary Computation (EC) community for two reasons: 1) As a simulation of evolutionary processes as they happen in nature, and 2) as a solution to hard optimization problems. With the passage of time EAs have become increasingly focused on function optimization. Given this narrowing of vision in the EC community, it is worth revisiting a paper written in 1997 by Hans-Paul Schwefel on the future challenges for EC. In that paper the author argues that the more an algorithm models natural evolution at work in the universe, the better it will perform (even in terms of function optimization). The present paper tests Schwefels hypothesis by designing an EA based on Charles Peirces theory of evolution. Peirces theory not only accounts for biological evolution on earth (as other theories of evolution do) but also offers an account of global, cosmological and universal evolution. In going beyond just biological evolution, Peirces theory of evolution meets the criteria suggested by Schewefel in his 1997 paper. The present paper mainly contributes in testing the Peircean EA on an extended set of benchmark optimization functions and compares the results with a classical EA that is based on Darwins theory of evolution. In majority of these comparisons the performance of the Peircean EA is notably superior. This exercise provides preliminary results that support Schwefels hypothesis. In return the experiments in evolutionary computation help provide important insights into Peirces theory of evolution.