Related Experiment Video
Updated: Nov 20, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
801
Design and Implementation of Fast Spoken Foul Language Recognition with Different End-to-End Deep Neural Network
Abdulaziz Saleh Ba Wazir1, Hezerul Abdul Karim1, Mohd Haris Lye Abdullah1
1Faculty of Engineering, Multimedia University, Cyberjaya 63100, Malaysia.
Sensors (Basel, Switzerland)
|January 26, 2021
Summary
Content censorship is vital for protecting young viewers from harmful language. This study introduces an intelligent system using deep learning (CNNs and RNNs) for effective profanity detection, outperforming existing methods with high accuracy and reduced computational cost.
Area of Science:
- Computer Science
- Artificial Intelligence
- Digital Media
Background:
- Manual content censorship of foul language is tedious and prone to errors due to human limitations.
- Excessive exposure to uncensored content negatively impacts character and behavior, especially in young viewers.
- Existing automated methods lack the robustness for comprehensive foul language detection.
Purpose of the Study:
- To develop an intelligent system for automated foul language censorship.
- To leverage advanced deep learning models for accurate profanity detection in digital media.
- To reduce the burden of manual content moderation and improve detection efficiency.
Main Methods:
- Collected, annotated, and augmented a dataset of foul language.
- Developed and evaluated deep Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) cells.
- Compared the performance of proposed models against state-of-the-art pre-trained neural networks.
Main Results:
- The proposed CNN and RNN systems demonstrated high accuracy in identifying curse words.
- Achieved a low False Negative Rate (FNR) ranging from 2.53% to 5.92%.
- Outperformed existing pre-trained models on a novel foul language dataset with reduced computational cost.
Conclusions:
- The proposed intelligent system is feasible and effective for foul language censorship.
- Deep learning models, specifically CNNs and RNNs (LSTM), offer a robust solution for automated profanity detection.
- The system provides a computationally efficient and accurate alternative to manual content moderation.
