4.6 Article

Regularized Urdu Speech Recognition with Semi-Supervised Deep Learning

Journal

APPLIED SCIENCES-BASEL
Volume 9, Issue 9, Pages -

Publisher

MDPI
DOI: 10.3390/app9091956

Keywords

speech recognition; locally linear embedding; label propagation; Maxout; low resource languages

Ask authors/readers for more resources

Automatic Speech Recognition, (ASR) has achieved the best results for English, with end-to-end neural network based supervised models. These supervised models need huge amounts of labeled speech data for good generalization, which can be quite a challenge to obtain for low-resource languages like Urdu. Most models proposed for Urdu ASR are based on Hidden Markov Models (HMMs). This paper proposes an end-to-end neural network model, for Urdu ASR, regularized with dropout, ensemble averaging and Maxout units. Dropout and ensembles are averaging techniques over multiple neural network models while Maxout are units in a neural network which adapt their activation functions. Due to limited labeled data, Semi Supervised Learning (SSL) techniques are also incorporated to improve model generalization. Speech features are transformed into a lower dimensional manifold using an unsupervised dimensionality-reduction technique called Locally Linear Embedding (LLE). Transformed data along with higher dimensional features is used to train neural networks. The proposed model also utilizes label propagation-based self-training of initially trained models and achieves a Word Error Rate (WER) of 4% less than that reported as the benchmark on the same Urdu corpus using HMM. The decrease in WER after incorporating SSL is more significant with an increased validation data size.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.6
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

Article Computer Science, Information Systems

LiMPO: lightweight mobility prediction and offloading framework using machine learning for mobile edge computing

Sardar Khaliq uz Zaman, Ali Imran Jehangiri, Tahir Maqsood, Nuhman ul Haq, Arif Iqbal Umar, Junaid Shuja, Zulfiqar Ahmad, Imed Ben Dhaou, Mohammed F. Alsharekh

Summary: The proliferation of mobile devices has led to the emergence of various services, but delivering task offloading results to users in the MEC environment is challenging, especially when user mobility is high. Traditional techniques handle computation offloading and mobility management separately, without considering real-time mobility factors, resulting in sub-optimal solutions. The LiMPO framework offloads compute-intensive tasks to user locations predicted by artificial neural networks and optimizes latency and energy consumption with a multi-objective genetic algorithm-based server selection technique.

CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS (2023)

Article Computer Science, Information Systems

Reinforcement learning for intelligent online computation offloading in wireless powered edge networks

Ehzaz Mustafa, Junaid Shuja, Kashif Bilal, Saad Mustafa, Tahir Maqsood, Faisal Rehman, Atta ur Rehman Khan

Summary: This article proposes a reinforcement learning-based intelligent online offloading framework, which can effectively make decisions between local or remote computation in a wireless powered MEC system, achieving optimal performance.

CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS (2023)

Article Computer Science, Information Systems

Multi-factor nature inspired SLA-aware energy efficient resource management for cloud environments

Sonia Bashir, Saad Mustafa, Raja Wasim Ahmad, Junaid Shuja, Tahir Maqsood, Abdullah Alourani

Summary: Cloud computing consumes a large amount of energy, leading to high expenditure, greenhouse gas emissions, and CO2 emissions. Existing energy-efficient techniques only consider the energy consumption of the CPU during task placement and ignore the energy consumption of memory and SLA violations. To address these issues, we propose two novel nature-inspired techniques based on artificial bee colony and particle swarm optimization, which consider the energy consumption of both CPU and memory during VM placement. We also provide SLA-aware variants to reduce SLA violations resulting from excessive task consolidation.

CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS (2023)

Article Engineering, Electrical & Electronic

Inspection of unmanned aerial vehicles in oil and gas industry: critical analysis of platforms, sensors, networking architecture, and path planning

Sana Nasim Karam, Kashif Bilal, Junaid Shuja, Faisal Rehman, Tahira Yasmin, Akhtar Jamil

Summary: Unmanned aerial vehicles (UAVs) have great potential in the oil and gas industry, especially in situations where human lives are at risk. They offer cost-effective and efficient monitoring solutions through carrying sensors and cameras. However, there are specific challenges to be addressed for the effective use of UAVs in the industry.

JOURNAL OF ELECTRONIC IMAGING (2023)

Article Computer Science, Hardware & Architecture

A methodology for shape matching of non-rigid structures based on integrated graphical information

Mingxuan Zhang, Muhammad Umair Hassan, Dongmei Niu, Xiuyang Zhao, Raheel Nawaz, Ibrahim A. Hameed, Saeed-Ul Hassan

Summary: This paper presents an automatic dense correspondence method for matching the mesh vertices of two 3D shapes under near-isometric and non-rigid deformations. The method combines three types of graphic structure information and includes three major steps: describing the vertices based on three types of graphical information, formulating the match as an optimization problem, and resolving the optimal solution using the projected descent optimization procedure. The method achieves superior performance to existing methods in quantitative and qualitative evaluations on challenging 3D shape matching datasets.

DISPLAYS (2023)

Article Computer Science, Artificial Intelligence

A transformer fine-tuning strategy for text dialect identification

Mohammad Ali Humayun, Hayati Yassin, Junaid Shuja, Abdullah Alourani, Pg Emeroylariffion Abas

Summary: Online medical consultation can improve the efficiency of primary health care. This paper proposes a fine-tuning strategy to identify the social origin of text authors, which can assist in selecting medical consultants for efficient communication. The proposed method achieves a 0.54% higher overall accuracy compared to the previous best result in the experiments.

NEURAL COMPUTING & APPLICATIONS (2023)

Article Physics, Multidisciplinary

A New Nonlinear Dynamic Speed Controller for a Differential Drive Mobile Robot

Ibrahim A. Hameed, Luay Hashem Abbud, Jaafar Ahmed Abdulsaheb, Ahmad Taher Azar, Mohanad Mezher, Anwar Ja'afar Mohamad Jawad, Wameedh Riyadh Abdul-Adheem, Ibraheem Kasim Ibraheem, Nashwa Ahmad Kamal

Summary: A disturbance estimation and rejection technique based on the improved active disturbance rejection control (IADRC) approach is proposed and verified on a ground two-wheel differential drive mobile robot. The IADRC is adopted to eliminate the effect of system uncertainties and external torque disturbance on both wheels. A novel nonlinear sliding mode extended state observer (NSMESO) is used to observe and cancel the generalized disturbance in real-time. Numerical simulations show a significant reduction in the ITAE index for both wheels, validating the efficacy of the proposed dynamic speed controller in damping the chattering phenomena and providing high insusceptibility to torque disturbance.

ENTROPY (2023)

Article Chemistry, Analytical

Predictive Maintenance of Norwegian Road Network Using Deep Learning Models

Muhammad Umair Hassan, Ole-Martin Hagen Steinnes, Eirik Gribbestad Gustafsson, Sivert Loken, Ibrahim A. A. Hameed

Summary: Industry 4.0 has revolutionized the use of physical and digital systems, especially in the digitalization of maintenance plans for physical assets. In this study, we developed a predictive maintenance approach using pre-trained deep learning models to effectively detect and classify different types of road damage. Our approach allows us to prioritize maintenance decisions based on the severity and occurrence of damage, providing a framework for efficient road maintenance. The evaluation of our proposed framework showed significant performance in various measures.

SENSORS (2023)

Article Multidisciplinary Sciences

Virtual Sensors for Nonlinear Discrete-Time Dynamic Systems

Oleg Sergiyenko, Alexey Zhirabok, Ibrahim A. Hameed, Ahmad Taher Azar, Alexander Zuev, Vladimir Filaretov, Vera Tyrsa, Ibraheem Kasim Ibraheem

Summary: This study investigates the problem of designing virtual sensors for nonlinear systems under disturbance. Two different mathematical techniques, algebra of functions and logic-dynamic approach, are used to solve the problem. The first technique provides a general solution, while the second technique uses linear algebra methods to find a solution specifically for nonlinear systems. The virtual sensors are designed to be robust against disturbance by utilizing invariant functions and estimating the prescribed function of the original system state vector. A practical example is provided to illustrate the theoretical results.

SYMMETRY-BASEL (2023)

Article Chemistry, Multidisciplinary

Plant-Inspired Soft Growing Robots: A Control Approach Using Nonlinear Model Predictive Techniques

Haitham El-Hussieny, Ibrahim A. Hameed, Ahmed B. Zaky

Summary: Soft growing robots, inspired by plant growth, excel in navigating tight and distant environments due to their flexibility and extendable lengths. However, controlling their tip position is challenging due to a lack of precise measurement methods. This paper proposes optimization-based approaches to achieve superior performance in point stabilization, trajectory tracking, and obstacle avoidance.

APPLIED SCIENCES-BASEL (2023)

Article Chemistry, Multidisciplinary

A New Approach to Nonlinear State Observation for Affine Control Dynamical Systems

Ahmad Taher Azar, Drai Ahmed Smait, Sami Muhsen, Moayad Abdullah Jassim, Asaad Abdul Malik Madhloom AL-Salih, Ibrahim A. Hameed, Anwar Ja'afar Mohamad Jawad, Wameedh Riyadh Abdul-Adheem, Vincent Cocquempot, Mouayad A. Sahib, Nashwa Ahmad Kamal, Ibraheem Kasim Ibraheem

Summary: In this paper, a Nonlinear Higher Order Extended State Observer (NHOESO) is proposed to replace the Linear Extended State Observer (LESO) in Conventional Active Disturbance Rejection Control (C-ADRC) solutions. The NHOESO extends the standard LESO by incorporating a two-term smooth nonlinear function with saturation-like characteristics. It allows for precise observation of generalized disturbances with higher-order derivatives. The stability of the NHOESO is analyzed using the Lyapunov method. Simulation results on an uncertain nonlinear Single-Input-Single-Output (SISO) system with time-varying external disturbances demonstrate the effectiveness of the proposed NHOESO in handling generalized disturbances compared to other ESOs.

APPLIED SCIENCES-BASEL (2023)

Article Green & Sustainable Science & Technology

An Interactive Multi-Criteria Decision-Making Approach for Autonomous Vehicles and Distributed Resources Based on Logistic Systems: Challenges for a Sustainable Future

Abduallah Gamal, Mohamed Abdel-Basset, Ibrahim M. Hezam, Karam M. Sallam, Ibrahim A. Hameed

Summary: The autonomous vehicle (AV) has the potential to restructure transportation infrastructure and improve traffic congestion, quality of life, and traffic safety. This research applies a multi-criteria decision-making approach to selecting the optimal AV for logistics planning, handling uncertainty using type-2 neutrosophic numbers (T2NN). Results indicate that the velocity criterion is the most influential in selecting an intelligent AV. Rating: 9/10

SUSTAINABILITY (2023)

Article Computer Science, Artificial Intelligence

An efficient and lightweight multiperson activity recognition framework for robot-assisted healthcare applications

Syed Hammad Hussain Shah, Anniken Susanne T. Karlsen, Mads Solberg, Ibrahim A. Hameed

Summary: Aging poses challenges to elderly individuals' social lives due to declining physical abilities, but group exercise in long-term care facilities is crucial for maintaining their physical and social well-being. However, accommodating these needs can be difficult due to staff shortages and lacking resources. To address this, a robotic exercise coach could be helpful. However, accurate and efficient human activity recognition is necessary for intelligent human-robot interaction in this context.

EXPERT SYSTEMS WITH APPLICATIONS (2024)

Article Computer Science, Artificial Intelligence

Deadline-aware heuristics for reliability optimization in ubiquitous mobile edge computing

Sardar Khaliq Uz Zaman, Tahir Maqsood, Azra Ramzan, Faisal Rehman, Saad Mustafa, Junaid Shuja

Summary: With the increasing demand for affordable and accessible broadband and mobile internet, the field of Ubiquitous Mobile Edge Computing (UMEC) has become highly dynamic. This study focuses on optimizing reliability in UMEC by considering latency and offloading failure probability. The proposed deadline-aware heuristic algorithm effectively reduces task failure ratio and achieves remarkable total latency, outperforming the state-of-the-art technique.

INTERNATIONAL JOURNAL OF DATA SCIENCE AND ANALYTICS (2023)

Review Computer Science, Hardware & Architecture

Leaf classification on Flavia dataset: A detailed review

Syed Umaid Ahmed, Junaid Shuja, Muhammad Atif Tahir

Summary: This paper examines commonly used and publicly accessible datasets for plant classification. Through the exploration of over 200 research papers, the advancements and developments in leaf classification, as well as new techniques and approaches, are discussed. The coherence and gaps in algorithms are highlighted for the benefit of future researchers.

SUSTAINABLE COMPUTING-INFORMATICS & SYSTEMS (2023)

No Data Available