Libyan Journal of Informatics

A Multimodal ASR System with Contextual Awareness and Emotional Sensitivity

Abeer Ali Aoun, Karim Dabbabi — Tue, 31 Dec 2024 00:00:00 +0000

The increasing demand for accurate speech recognition systems in diverse languages, particularly Arabic, poses significant challenges due to variations in dialects, background noise, and emotional context. Traditional Automatic Speech Recognition (ASR) models often struggle to maintain high accuracy in the presence of these factors, leading to suboptimal performance in real-world applications. This study presents a novel Multimodal ASR system that addresses these challenges by integrating audio, visual, and emotional cues to enhance both transcription accuracy and emotion detection for Arabic speech.

The proposed model was evaluated on the Audio-Visual Arabic Natural Emotion (AVANEmo) dataset, employing state-of-the-art techniques, including Wav2Vec 2.0 for audio feature extraction, convolutional neural networks for lip movement recognition, and a contextual language model to refine outputs. The system achieved a Word Error Rate (WER) of 16.3% and a Character Error Rate (CER) of 10.7%, outperforming existing models such as DeepSpeech (19.4% WER, 13.7% CER) and Jasper (18.2% WER, 12.9% CER). Moreover, the proposed model demonstrated a notable accuracy of 88.9% for emotion detection, significantly surpassing the performance of previous models, which reported 84.2% accuracy. These results underscore the efficacy of the multimodal approach in enhancing Arabic speech recognition and emotion classification, highlighting its potential for real-world applications.

A Semantic Based Gender Identifications through User Generated Contents

Mohammed Ali Ibrahim Eltaher — Tue, 31 Dec 2024 00:00:00 +0000

User gender is crucial information for personalized services and applications in online social networks. It impacts areas such as recommendation systems, advertising, and connection discovery. However, user gender information may be hidden or not specified in online social networks, leading to inaccuracies or limitations in various applications. The daily interactions of billions of users on online social networks like Flickr contribute to creating vast amounts of user-generated content. This content includes multiple media such as images, videos, and textual information. The primary aim of this paper is to address the challenge of identifying the gender of users. Our approach involves a semantic-based data technique. Using a semi-automatic image tagging system implies a process where images are labeled or categorized with automation, potentially improving efficiency and accuracy. We employ two classification algorithms for gender identification: Naive Bayes and Support Vector Machines (SVM), where data are typically represented as feature vectors. Our experimental results on more than 149,700 Flickr users demonstrate an accuracy of over 84% for gender identification. This suggests that combining Naive Bayes and SVM algorithms, with data represented as feature vectors, has proven effective in classifying gender based on user-generated content.

Assessing Queue Management Strategies to Enhance Quality of Service in MPLS VPN Networks

Mahmud Mansour, Ahmed Samood, Najia Ben Saud — Tue, 31 Dec 2024 00:00:00 +0000

Abstract. Multiprotocol Label Switching (MPLS) has emerged as a key technology for providing quality of service (QoS) guarantees in IP networks. This paper presents an extensive simulation-based evaluation of MPLS QoS mechanisms. Specifically, different queuing policies - First In First Out (FIFO), Priority Queuing (PQ), and Weighted Fair Queuing (WFQ) are implemented and analyzed for providing end-to-end QoS for real-time voice and data applications over an MPLS VPN backbone. The network simulations are performed using the OPNET tool with detailed MPLS, VPN, and queuing parameters configuration. Performance is evaluated across multiple metrics including jitter, delay variation, end-to-end delay, traffic sent/received for both voice and data flows. The results demonstrate that FIFO queuing delivers the best QoS performance for voice traffic, providing simple first-in, first-out buffering. WFQ is shown to outperform PQ for voice flows. All queuing mechanisms can meet QoS requirements for voice and data applications. The paper provides a comprehensive investigation into configuring MPLS networks with QoS capabilities. Key findings show that MPLS VPNs effectively reduce network complexity and costs. FIFO emerges as an optimal queuing technique for enabling QoS services in MPLS networks carrying multimedia applications.

Performance Study of ETX Metric in Flight Ad-Hoc Networks

Mohamed Alrayes, Zayed Khalifa — Tue, 31 Dec 2024 00:00:00 +0000

Flying Ad-hoc networks (FANETs) are wireless networks that allow unmanned aerial vehicles (UAVs) to interact with one another without a permanent infrastructure. For effective data transfer in FANETs, routing protocols like Ad hoc On-Demand Distance Vector (AODV) are crucial.

While several studies have investigated the use of ETX (Expected Transmission Count) in specialized networks such as Vehicular Ad-hoc Networks (VANETs) and mesh networks, there is a lack of comprehensive study across routing protocols under FANETs.

This research gap shows how important it is to have a structured way to check how well ETX
metrics work in FANETs, taking into account things taking into account network structure,
mobility patterns, and environmental circumstances.

The research examined the efficacy of ETX and hop count routing metrics utilizing the NS-3 network simulator. In this research, simulations of Ad hoc On-Demand Distance Vector (AODV) routing protocol were conducted in two scenarios: one employing the ETX metric (AODV-ETX) and the other utilizing the conventional hop count metric (original AODV). Various performance metrics, including average throughput, average end-to-end delay, packet delivery ratio (PDR), and useful traffic ratio (UTR), are evaluated.

The results show that in varying density of network environments, the ETX metric leads to higher delays but improves packet delivery ratios, throughput, and UTR. However, in scenarios with varying speeds of UAVs, it leads to a high end-to-end delay, a lower packet delivery ratio, lower throughput, and a lower UTR, compared with the hop-count metric. It is observed as well that the ETX metric performed better than the hop count metric in terms of end-to-end delay, packet delivery ratio, throughput, and UTR when the network load changes.

Automatic Verb Detection in Libyan dialect

Abdusalam nwesri, Nabila Almabrouk S. Shinber — Tue, 31 Dec 2024 00:00:00 +0000

Automatic recognition of verbs is crucial to a wide range of natural language processing tasks. Verbs exhibit the relational information in a sentence between the action and its participant and are considered the primary source of information in understanding a sentence and the base for any NLP task. In this paper, we experiment six machine learning algorithms to identify verbs from other words in the Libyan dialect. Among algorithms used, the Support vector classifier (SVC) was best at identifying verbs with a micro F1 score of 70%.