FOOD ENGINEERING, EXTRACTION METHODS, FOOD SAFETY AND STANDARDS
6
Scopus Publications
Scopus Publications
An Offline Mixed-Language Voice-Driven Billing Assistant Using Vosk Speech Recognition Swathi K, Rampradop S, Vaseem Akthar A, Mathankumar S 2025 2nd International Conference on Intelligent Technologies for Sustainable Electric and Communications Systems Itech Secom 2025, 2025 Most existing voice-based applications depend on cloud services to process speech, which means they require continuous internet connectivity. This dependency often causes problems such as slow response times, high data usage, recurring subscription costs, and even complete failure in areas with poor or no network coverage. To overcome these limitations, this project introduces an offline Android assistant that supports both billing and learning tasks through Tamil-English voice commands. The system is powered by the Vosk Speech Recognition Toolkit, which allows it to accurately recognize commands without relying on the internet. After recognizing the input, the app provides immediate spoken feedback in both Tamil and English so that users can easily confirm their actions. It also generates professional invoices in PDF format, making it suitable for small shopkeepers who want to maintain digital records. For security, the application includes a voice authentication feature that verifies the user before granting access. All data, including past billing records, is stored locally using the Room Database, ensuring persistence even if the device restarts or goes offline.
Enhanced Semantic Duplicate Question Detection using MPNet based Sentence-BERT Swathi K, Kanishkha R, Kisanth S, Krishnasamy R Proceedings of 5th International Conference on Ubiquitous Computing and Intelligent Information Systems Icuis 2025, 2025 Web-Based question-and-answer (Q&A) websites like Quora and Reddit suffer the chronic issue of repeated questions, which in turn results in redundancy, lower quality of content, and bad user experience. Deeper semantic relevance cannot be captured by traditional approaches that rely on lexical similarity, especially when dealing with brief, informal, or context-specific questions. To address this, we present a system that more precisely detects duplicate questions by using Sentence-Bert embeddings in conjunction with cosine similarity scoring. After the model was adjusted and trained using benchmark datasets, Streamlit was used to create the application-based web user interface and the research testing terminal-based interface. In order to perform real-time duplicate checking over live user submissions, we also integrated Reddit APIs, making the system dynamic and application-ready. According to experimental testing, our approach balances recall and accuracy by achieving a threshold similarity of 0.69 at optimal classification. This paper pushes duplicate detection to real-time community-driven platforms while providing a lightweight yet powerful alternative to deep hybrid models.
Emotion Analysis of English-Translated Tamil Literature K Swathi, G K Kamalam, N Suganya Baby, A Aadhishri, D K Kawethaa Shree 2024 15th International Conference on Computing Communication and Networking Technologies Icccnt 2024, 2024 Focus of this study is to investigate the nuanced world of emotions in English-translated Tamil literature through the lens of sentiment analysis. The dataset consists of poems expressing a range of emotions including joy, sadness, love, anger, pride, courage, and fear. To effectively analyze and classify these emotions, we utilized cutting-edge transformer-based models, like BERT, DistilBERT, and RoBERTa, to conduct our analysis. The outcomes of our study showcase the exceptional capabilities of these models in effectively analyzing emotions within poetic expressions. RoBERTa, in particular, emerged as a standout choice, consistently achieving high accuracy, precision, recall, and $F 1$ scores across all emotion categories. Its extensive pre-training and dynamic masking techniques allowed itto excel in capturing the intricate emotional nuancesembedded in the translated poems. DistilBERT demonstrated commendable efficiency without compromising accuracy, making it a practical option for real-time sentiment analysis. BERT, although a strong performer, showcased marginally reduced efficiency compared to RoBERTa and DistilBERT.
Multilingual Exploration of the Kambaramayana: Analyzing Tamil Translations in Telugu, Malayalam, and Kannada Using mBERT and DistilBERT Swathi K, Kamalam G K, Hemalatha S, Jothimani K, Dharshini N, Aadhishri A Proceedings IEEE 2024 1st International Conference on Advances in Computing Communication and Networking Icac2n 2024, 2024 An essential and basic aspect of our existence are our emotions. Our behaviors and words reflect our feelings, even though they are not instantly obvious. This research looks at the multilingual features of the Kambaramayana, with a concentration on Tamil translates into Telugu, Malayalam, and Kannada. We study how linguistic and cultural subtleties show across translations using a varied dataset. We do a comparison study using sophisticated multilingual language models, specifically mBERT and DistilBERT, to identify semantic variances and similarities. We hope to uncover the subtle interplay among language and culture in literary text transmission by investigating vocabulary, grammar, and cultural allusions. Our findings not only improve the accuracy of cross-lingual text analysis, but also highlight the Ramayana's lasting relevance among South Indian language communities. The experiments clearly reveal that mBERT outperforms DistilBERT in emotion analysis tasks between languages.
Multimodal Approach for depression detection: Integrating speech and eye ball movement data Hemalatha S, Jothimani K, Swathi K, Shibinta S, Jason Selvakumar W, Sathish D Proceedings IEEE 2024 1st International Conference on Advances in Computing Communication and Networking Icac2n 2024, 2024 Depression is a common mental health condition that impacts millions of people globally, its diagnosis often relies on subjective assessments. This project introduces a Multimodal Convolutional Neural Network (MCNN) framework tailored for depression detection, leveraging speech and eye movement data as input modalities. The project's foundation lies in combining advanced deep learning techniques with multimodal data analysis. Through the MCNN architecture, the system learns to extract discriminative features from speech recordings and eye movement patterns, enabling holistic assessments of individuals' mental health status. Data acquisition involves the collection and preprocessing of multimodal data samples from individuals both with and without depression. The MCNN model is trained on this curated dataset, exploiting the convolutional layers to process spatial features from eye movement images and temporal features from speech spectrograms concurrently. The data set is preprocessed using different algorithms of MCNN. We use MCFF’s algorithm, which represents the spectral envelope of the speech signal. Eyeball movement is processed by cluster-based algorithms that Convert raw eye movement data into a suitable format for analysis. This may involve segmenting the data into fixations and saccades, and representing them as sequences of gaze positions or heatmaps. The anticipated outcome of this project is a scalable, efficient, and accurate system for depression detection, capable of outperforming conventional diagnostic methods. Future directions include extending the MCNN framework to incorporate additional modalities, such as physiological data, and deploying the system in real-world clinical settings to facilitate early intervention and personalized treatment strategies.
