Adil Bakri

Packet loss in Voice over Internet Protocol (VoIP) communications can significantly degrade speech quality, particularly affecting Conjugate-Structure Code-Excited Linear Prediction (CS-CELP) type speech coders like G.729. Traditional Packet Loss Concealment (PLC) methods often struggle with consecutive frame losses. Recent advancements in deep learning have introduced Deep Neural Networks (DNNs) as a promising solution for PLC. This article explores the integration of DNNs into PLC for CS-CELP-type speech coders, focusing on methodologies, experimental setups, results, and future directions.

This paper presents the design and implementation of a voice-controlled robotic system that uses Arabic Automatic Speech Recognition (ASR) to assist people with special needs, particularly those with mobility disabilities. Two hardware prototypes were developed: (1) a voice-controlled LED lighting circuit and (2) a miniature voice-controlled robotic vehicle. Both systems employ the AMR Voice Android application, an Arduino microcontroller, and an HC-05 Bluetooth module to convert Arabic spoken commands into executable control signals in real time. Experiments were conducted with 10 native Arabic speakers (5 male, 5 female) using 6 isolated command words (120 total trials). The system achieved an overall recognition accuracy of 92.5%, with no statistically significant difference between male and female speakers. Compared to existing English-only assistive systems, our work makes a distinct contribution by targeting Modern Standard Arabic and demonstrating deployability on low-cost, off the shelf hardware. These results confirm the system's viability for real-world assistive applications in Arab communities.