Waleed H. Abdulla

Cross-words reference template for DTW-based speech recognition systems

One of the main problems in dynamic time-warping (DTW) based speech recognition systems are the preparation of reliable reference templates for the set of words to be recognised. This paper presents a simple novel technique for preparing reliable reference templates to improve the recognition rate score. The developed technique produces templates called crosswords reference templates (CWRTs). It extracts the reference template from a set of examples rather than one example. This technique can be adapted to any DTW-based speech recognition systems to improve its performance. The speaker-dependent recognition rate, as tested on the English digits, is improved from 85.3%, using the traditional technique to 99%, using the developed technique.

Hardware–Software Codesign of Automatic Speech Recognition System for Embedded Real-Time Applications

We present a hardware-software coprocessing speech recognizer for real-time embedded applications. The system consists of a standard microprocessor and a hardware accelerator for Gaussian mixture model (GMM) emission probability calculation implemented on a field-programmable gate array. The GMM accelerator is optimized for timing performance by exploiting data parallelism. In order to avoid large memory requirement, the accelerator adopts a double buffering scheme for accessing the acoustic parameters with no assumption made on the access pattern of these parameters. Experiments on widely used benchmark data show that the real-time factor of the proposed system is 0.62, which is about three times faster than the pure software-based baseline system, while the word accuracy rate is preserved at 93.33%. As a part of the recognizer, a new adaptive beam-pruning algorithm is also proposed and implemented, which further reduces the average real-time factor to 0.54 with the word accuracy rate of 93.16%. The proposed speech recognizer is suitable for integration in various types of voice (speech)-controlled applications.

Fast track article: Ambient intelligence platform using multi-agent system and mobile ubiquitous hardware

In this paper, a novel ambient intelligence (AmI) platform is proposed to facilitate fast integration of different control algorithms, device networks and user interfaces. This platform defines the overall hardware/software architecture and communication standards. It consists of four layers, namely the ubiquitous environment, middleware, multi-agent system and application layer. The multi-agent system is implemented using Java Agent DEvelopment (JADE) framework and allows users to incorporate multiple control algorithms as agents for managing different tasks. The Universal Plug and Play (UPnP) device discovery protocol is used as a middleware, which isolates the multi-agent system and physical ubiquitous environment while providing a standard communication channel between the two. An XML content language has been designed to provide standard communication between various user interfaces and the multi-agent system. A mobile ubiquitous setup box is designed to allow fast construction of ubiquitous environments in any physical space. The real time performance analysis shows the potential of the proposed AmI platform to be used in real-life AmI applications. A case study has also been carried out to demonstrate the possibility of integrating multiple control algorithms in the multi-agent system and achieving a significant improvement on the overall offline learning performance.

Effects of Imperfect Secondary Path Modeling on Adaptive Active Noise Control Systems

Implementation of adaptive active noise control (ANC) systems requires an estimate model of the secondary path to be uploaded onto digital control hardware. In practice, this model is not necessarily perfect; however, to avoid mathematical difficulties, theoretical analysis of these systems is usually conducted for a perfect secondary path model. This paper conducts a stochastic analysis on performance of Filtered-x LMS (FxLMS)-based ANC systems when the actual secondary path and its model are not identical. This analysis results in a number of mathematical expressions, describing effects of a general secondary path model on stability, steady-state performance and convergence speed of FxLMS-based ANC systems. As a surprising result, it is found that intentional misadjustment of secondary path models can enhance performance of ANC systems in practice. Theoretical results are found to be in a good agreement with the results obtained from numerical analysis. Also, experimental results confirm the validity and accuracy of the theoretical results.

On the convergence of real-time active noise control systems

Available convergence analyses of adaptive active noise control systems apply to only theoretical cases with broad-band white noise or pure delay secondary paths. In order to investigate convergence behaviors of these systems in more practical conditions, this paper conducts a new convergence analysis for filtered-x LMS-based active noise control systems with band-limited white noise and moving average secondary paths. A linear model for the adaptation process is developed. Based on this model, the upper-bound of the adaptation step-size is derived. Also, the adaptation step-size leading to the fastest convergence rate is derived. In addition to the computer simulation, a fully implemented real time active noise control system is used to verify the validity of the analytical results.

A software algorithm prototype for optical recognition of embossed Braille

Braille is a tactile format of written communication for sight-impaired people worldwide. This paper proposes a software solution prototype to optically recognise single sided embossed Braille documents using a simple image processing algorithm and probabilistic neural network. The output is a Braille text file formatted to preserve the layout of the original document which can be sent to an electronic embosser for reproduction. Preliminary experiments have been performed with an excellent recognition rate, where the transcription accuracy is at 99%.

On the stability of adaptation process in active noise control systems.

The stability analysis of the adaptation process, performed by the filtered-x least mean square algorithm on weights of active noise controllers, has not been fully investigated. The main contribution of this paper is conducting a theoretical stability analysis for this process without utilizing commonly used simplifying assumptions regarding the secondary electro-acoustic channel. The core of this analysis is based on the root locus theory. The general rules for constructing the root locus plot of the adaptation process are derived by obtaining root locus parameters, including start points, end points, asymptote lines, and breakaway points. The conducted analysis leads to the derivation of a general upper-bound for the adaptation step-size beyond which the mean weight vector of the active noise controller becomes unstable. Also, this analysis yields the optimum step-size for which the adaptive active noise controller has its fastest dynamic performance. The proposed upper-bound and optimum values apply to general secondary electro-acoustic channels, unlike the commonly used ones which apply to only pure delay channels. The results are found to agree very well with those obtained from numerical analyses and computer simulation experiments.

A preprocessing algorithm for handwritten character recognition

Abstract This paper describes an efficient algorithm for extracting a simplified skeletal version of hand-written characters. In this version, curvatures are represented by straight lines connecting tree-, edge- and endpoints in the right sequence. These three basic feature points are detected using an efficient algorithm.

Reduced feature-set based parallel CHMM speech recognition systems

This paper presents the multi-streams paradigm as a technique for improving speech signal feature set design and as a performance booster for speech recognition systems, based on the continuous-density hidden Markov model (CHMM) framework. In the multi-streams paradigm we are dealing with different feature sets independently to estimate the same task, and then combining their results at a suitable stage. This paradigm combines the strengths of many varied feature vectors to attain better statistical estimation. Under the proposed paradigm the feature vectors are split into three independent streams, and each stream is used to model an independent CHMM. Then the outcomes of these models, when subjected to any speech input, are merged under a certain strategy. This technique alleviates the dominance effect of the features, and reduces the dimensionality of the feature vectors used in each model. The F-ratio technique is used to further reduce the dimensionality of each stream. Experimental results on different datasets show superiority of the developed paradigm over the corresponding single-stream baseline.

A secure and robust audio watermarking scheme using multiple scrambling and adaptive synchronization

In this paper, a novel highly confidential audio watermarking scheme using multiple scrambling is presented. Superior to other techniques, the new scheme is self-secured by integrating multiple scrambling operations into the embedding stage. To ensure that unauthorized detection without correct secret keys is nearly impossible, the following have been made. The watermark is encrypted by a coded-image; certain subbands are randomly selected from the total subbands for embedding and their order of coding is further randomised. In addition, adaptive synchronization is used to improve the robustness against hazardous synchronization attacks, such as random samples cropping/inserting, pitch-invariant time stretching and tempo-preserved pitch shifting. Our goal is to make the watermark impossible to be detected and robust even though the watermarking algorithm is open to the public. Experimental results of listening evaluation and robustness tests have demonstrated that the proposed scheme preserves transparent perception and also possesses strong resistance to typical attacks on audio watermarking.