Ch. Kalyani1, K. V. K. Saibhaskar2, K. Lavanya Lakshmi2, K. E. S. R. K. Raju2, Rehana Tabassum2, V. Sai Gowtham2, 1Assistant Professor, Computer Science and Engineering, ANITS, Visakhapatnam, Andhra Pradesh, India, 2Computer Science and Engineering, ANITS, Visakhapatnam, Andhra Pradesh, India
Sign language, which allows for effective expression through gestures and movements, is one of the most crucial forms of communication for those who are deaf or hard of hearing. However, tools for translating spoken or written language into Indian Sign Language (ISL) are limited. This paper introduces SignSynth, a novel system designed to convert audio, text, and video content into ISL videos. By leveraging speech-to-text conversion, natural language processing, and tokenization, the system processes meaningful words while omitting redundant terms. These words are mapped to a curated database of ISL gestures represented as videos, which are then merged into a seamless output using advanced video processing techniques. With over 63 million individuals in India experiencing hearing disabilities, SignSynth offers a scalable and cost-effective solution to bridge communication gaps. This system has significant potential to enhance accessibility and facilitate interaction between hearing and non-hearing individuals, paving the way for advancements in real-time sign language translation
Indian Sign Language, Speech-to-Text, Gesture Recognition, Neural Networks, Sign Language Translation.
Balasubramanian S1 and Natarajan M2, 1Research Scholar, Department of Computer and Information Science, Annamalai University, Tamil Nadu, India,2Assistant Professor/Programmer, Department of Computer and Information Science, Annamalai University, Tamil Nadu, India
Due to ever-changing market conditions, predicting crude oil prices is complex and challenging. Traditional models struggle to adapt, while advanced AI models like LSTMs can find hidden patterns but often lack transparency. This study compares Machine Learning (Random Forest) and Deep Learning (LSTM) to assess both accuracy and interpretability. It also explores Explainable AI (XAI) methods like SHAP and LIME to make AI predictions more understandable for policymakers. The results show that combining machine learning, deep learning, and XAI techniques improves both the accuracy and trustworthiness of crude oil price forecasts.
Explainable AI, Trade Forecasting, Random Forest, LSTM, SHAP, LIME, Machine Learning, Deep Learning, xAI, Crude Oil Trade.
Kundanika Pradhan and Muhammed Anish, Department of Computational Technologies, SRM University, Tamil Nadu, India
Guaranteeing consumer safety and regulatory compliance for the cosmetics and personal care market involves thoroughly examining product ingredients. Conventional methods proved inadequate in detecting hazardous materials accurately and with high efficiency. The OCR retrieves ingredient information from product labels and packaging and then analyzes it through ChemBERTa, a transformer model trained on chemical representations. A specially developed module retrieves the SMILES representation of each ingredient extracted via an API-based mechanism. ChemBERTa is compared with the usual machine learning classifiers such as Support Vector Machine (SVM), Random Forest, Decision Trees, Bagging, and XGBoost in this work. ChemBERTa is a better option than traditional classifiers, and it has better accuracy when predicting the carcinogenicity of chemical compounds. In addition, an easy-to-use interface has been created with Streamlit that combines ChemBERTa and Llama 3.2 to present an easy and informative experience for users.
oncogenic chemicals, deep learning, ChemBERTa, optical character recognition, classification model, chemical informatics, Streamlit, Llama 3.2.