THE CSI JOURNAL ON COMPUTER SCIENCE AND ENGINEERING
Year:2007 | Volume:5 | Issue:1 (A)|Papers Count:5

Writer Identification recently has been studied and it has a wide variety of applications. Most Studies about writer identification are based on English documents and to our knowledge no research has been reported on Farsi or Arabic documents. In this paper we have proposed a method for off-line writer Identification which, text-independent. Based on the idea that has been presented in the previous studies here we assume handwriting as texture image and a set of features which are based on multi-channel Gabor filters are extracted Horn preprocessed image of documents.Substantially, the property of proposed method is using of the bank of gabor filters which is appropriate for structure of farsi handwritten texts and vision system. Our method with two methods which are based on co-occurrence matrix and gabor filters, are implemented and experimental results on handwriting of 25 peoples demonstrate that the proposed method achieves better performance on farsi handwritten documents.

In this paper SYS-PNLMS algorithm is proposed. The analysis reveals that it performs a faster convergence rate compared to that of the recently introduced SPNLMS, PNLMS algorithms. Compared with its proportionate counterparts e.g. PNLMS and SPNLMS, the proposed SYS-PNLMS algorithm not only results in a faster convergence rate for both white and colored noise inputs, but also preserves its initial fast convergence rate until it reaches to its steady state condition. It also presents a higher tracking behavior for quasi-stationary inputs such as speech signal in addition to better performance in terms of computational complexity and resulting ERLE. In addition, the proposed SYS-PNLMS algorithm is also evaluated with previously proposed algorithms in a theoretical framework which validates the computer simulation results in terms of CPU time and number of iterations needed for each algorithm to get converged. Finally, a region of convergence for the proposed algorithm is derived for different input cases including white, colored noise and speech signal. This region is also compared with the practical value usually used in echo cancellation application.

In recent years, the unit selection-based concatenative speech synthesis method using a large corpus has attracted great attention as it produces more natural output speech compared to other known approaches. Also this method has a great potential for improvement, even its base idea seems to be simple. The main components of this technique include a corpus containing variant instances, two criteria, namely target cost and concatenation cost, for evaluation of the instances, and finally a search algorithm for identification and selection of the best instances. In this paper, we present the structure of proposed unit selection synthesis system for Farsi language that is entitled FarsBayan. In this research, the constitutive sub-costs of cost measures, the different methods for determining sub-cost weights and pruning algorithms to reduce search space are described. The output speech was found to be remarkably fluent and natural. The quality of the output speech has been evaluated using MOS subjective test, and we have obtained a MOS test score of 3.8 for overall quality.

Generating pronunciation variants of words is an important applicable subject in speech research and is used extensively in automatic speech recognition and segmentation systems. In this way, decision trees are extensively used to model pronunciation variants of words and sub-word units. In case of word units and very large vocabulary, in order to train necessary decision trees, a huge amount of speech utterances that contain all of the needed words in vocabulary with a sufficient number of repetitions for each one is required; additionally an extra corpus is needed for every word Which is not included in the original training corpus and may be added to the vocabulary In the future. To solve these drawbacks, we have designed generalized decision trees, which can be trained using a medium-size corpus over group’s similar words to share information on pronunciation, instead of training a separate tree for ever single word. Generalized decision trees predict places in word where substitution, deletion and insertion phonemes may occur. Next to this step, In order to specifically determine word variants appropriate statistical contextual rules are applied to the permitted places. The hybrids of generalized decision trees and contextual rules are designed in static and dynamic versions. The hybrid static pronunciation models take into account word phonological structures, unigram probabilities, stress and phone context information simultaneously, while the hybrid dynamic models consider an extra feature. Speaking rate 10 generate pronunciation variants 0f words. Using the word variants, generated by static and dynamic models, in the lexicon of SHENAVA Persian continuous speech recognizer, relative word error raye reductions of as high as 8.1% and 10.3% are obtained respectively.

WA*, which has been widely used in computer science and AI applications is a search method based on A*. This method does not guaranty to find optimal solutions, but with the appropriate determination of the weights, nearly optimal solutions can be efficiently found. Generally, the systems that use WA* as their search mechanism, use a constant weight for all problems in various domes, while different domains have different characteristics. In this paper, we show that using a constant weight for all domains and problems is inappropriate. We also introduce a novel search strategy named GBFS* as an upper bound of the WA* performance.