My Publications

Google Scholar Profile

Below is a collection of my published works, highlighting my contributions to the fields of research computer science, artificial intelligence, and related areas. These publications reflect my ongoing exploration and commitment to advancing knowledge and innovation in these domains. Feel free to explore them to learn more about my work and insights.

1. Exploring the Power of Deep Learning for Seamless Background Audio Generation in Videos

Published: 2022
Journal: IEEE Access
Authors: Anjana S. Nambiar, Kanigolla Likhita, K. V. S. Sri Pujya, Niharika Panda
DOI: 10.1109/ICCCNT56998.2023.10306607

Overview:
This research explores the integration of deep learning models to automate the generation of background audio in videos, aiming to enhance the immersive quality and contextual coherence of multimedia content. The approach leverages temporal and spatial visual cues to synthesize audio components that align naturally with visual scenes.

Key Contributions:

Designed deep learning pipelines that analyze video features to generate background audio dynamically.
Achieved an accuracy of 89% using Temporal Segment Networks (TSN) trained over 30 epochs.
Demonstrated applicability in domains such as audio restoration, event detection, and AI-based video editing.
Discussed future improvements for model efficiency and adaptability across diverse media genres.

Impact:
This work contributes toward advancing tools for intelligent media processing and has potential applications in virtual content creation, automated dubbing, and accessibility for individuals with visual impairments.

2. Comparative Study of Deep Classifiers for Early Dementia Detection Using Speech Transcripts

Published: 2022
Journal: IEEE Access
Authors: Anjana S. Nambiar, Kanigolla Likhita, K. V. S. Sri Pujya, Deepa Gupta, Susmitha Vekkot, S. Lalitha
DOI: 10.1109/INDICON56171.2022.10039705.

Overview:
This study presents a comprehensive analysis of various deep learning models for detecting early signs of dementia using English speech transcript data. The goal was to identify efficient model architectures capable of capturing linguistic patterns correlated with cognitive decline.

Key Contributions:

Employed multiple embeddings, including GloVe, Word2Vec, Doc2Vec, and Transformer-based models such as BERT, RoBERTa, and ALBERT.
Combined embedding layers with sequential models like LSTM, BiLSTM, and GRU.
Evaluated performance on the Pitt Corpus from the DementiaBank dataset.
Achieved:
- 81.2% accuracy with the BERT + BiLSTM model.
- 0.81 F1-score with ALBERT + BiLSTM.

Impact:
The results support the feasibility of deep learning in clinical linguistics and healthcare, enabling development of AI-driven tools for non-invasive, early-stage dementia screening.

3. Design of Super Mario Game Using Finite State Machines

Published: October 2022
Conference: Computer Networks and Inventive Communication Technologies
Authors: Anjana S. Nambiar, Kanigolla Likhita, K. V. S. Sri Pujya, M. Supriya
DOI: https://doi.org/10.1007/978-981-19-3035-5_55

Overview:
This paper illustrates how finite state machines (FSMs) can be utilized to simulate interactive behaviors in games, focusing on character logic within the Super Mario universe. The project demonstrates the application of formal computation models in modern game design.

Key Contributions:

Modeled character behaviors and transitions (e.g., movement, power-ups, damage) through FSMs.
Developed the game's logic using Python and visualized state diagrams using JFLAP.
Presented a structured approach for integrating automata theory into interactive storytelling and game logic.
Addressed challenges in real-time state transition handling and player feedback mechanisms.

Impact:
This work contributes to educational and developmental fields by connecting computational theory with practical implementation, serving as a reference model for academic instruction in computer science and game development.