About
Dr. Kruthika S. G is an Assistant Professor at the School of Engineering, RV University, Mysuru Campus, with over 11 years of experience in teaching, research, industry collaborations, and academic administration. She holds a Ph.D. in Computer Science and Engineering with specialization in Digital Forensics from JSS Science and Technology University, Mysuru.
Her academic and research interests span Artificial Intelligence, Digital Forensics, Cyber Security, Speech and Image Processing, Deep Learning, and Web Technologies. Dr. Kruthika is dedicated to fostering a research-driven, outcome-based, student-centric learning environment that bridges classroom knowledge with industry requirements and real-world applications.
She has published more than 15 research articles in Scopus- and Web of Science-indexed journals, including Q1 and Q2 publications. In addition, she has contributed three international book chapters with reputed publishers such as Wiley, Springer, and Taylor & Francis CRC Press. Dr. Kruthika holds two published Indian patents. Her current research pipeline includes additional book chapters, conference papers, and journal manuscripts that are under review.
Dr. Kruthika is the recipient of the DST WISE-KIRAN Women Scientist Fellowship and the Best Paper Presentation Award. She has contributed to research proposals supported by DST, VGST, and AICTE, and serves as a reviewer for IEEE, Scopus-indexed journals, and international conferences. She is also an invited resource person for sessions on Artificial Intelligence, Digital Forensics, and Cyber Security.
Along with her teaching and research activities, she has led AICTE ATAL workshops, faculty development programs, hackathons, student outreach initiatives, and academic events. Her institutional contributions include NBA documentation, faculty mentoring, academic planning, proctoring, inventory management, and coordination. She serves as a member of the Board of Examination at JSS STU and is a lifetime member of ISTE.
Semi-Automatic Voice Comparison Approach Using Spiking Neural Network for Forensics (2024):
This patent, authored by Kruthika S.G., Trisiladevi C. Nagavi, and P. Mahesha, introduces a novel **semi-automatic voice comparison framework** specifically designed for forensic applications. Published in April 2024, the invention leverages **Spiking Neural Networks (SNNs)**, which are biologically inspired neural models that process information using discrete events or "spikes". This approach is likely intended to improve the temporal processing of voice signals, which is critical in forensic investigations where identifying a suspect with high confidence is necessary. SNNs are often preferred for their energy efficiency and ability to handle time-series data more effectively than traditional artificial neural networks. The development of this patent is a key output of the author’s DST-funded project aimed at advancing forensic speaker verification through high-accuracy AI-driven algorithms.
Improving Speaker Verification in Forensics Using DBN-Based Representation Learning (2025):
This patent, filed in September 2025, focuses on enhancing the accuracy of speaker verification systems through **Deep Belief Networks (DBN)**. Deep Belief Networks are generative graphical models that are particularly effective at learning hierarchical representations of data. In the context of forensic voice comparison, this invention utilizes DBNs to learn deeper, more robust features from voice samples that are less susceptible to environmental noise or voice disguise. By improving the representation of speaker-specific traits, this system aims to provide forensic experts with more reliable evidence for legal proceedings. This patent represents a significant advancement in the author's research into hierarchical deep learning models and their specific utility in verifying identities through vocal characteristics under challenging forensic conditions.
Link: https://search.ipindia.gov.in/IPOJournal/Journal/ViewJournal
Voice Comparison Using Acoustic Analysis and Generative Adversarial Network for Forensics (2025):
Published in the *International Journal of Image, Graphics and Signal Processing*, this paper combines traditional **acoustic analysis** with **Generative Adversarial Networks (GANs)** for forensic voice comparison. The research likely utilizes GANs to synthesize or augment voice data, helping to train more robust models even when suspect samples are limited or of poor quality. By integrating acoustic features—such as pitch, tone, and resonance—with the generative capabilities of GANs, the study offers a dual-layered approach to identifying speaker characteristics. This research is significant because it addresses the data scarcity issues common in forensic work, where high-quality recordings of suspects are often difficult to obtain. The use of GANs allows for the creation of more resilient identification systems that can operate effectively across various recording environments. DOI: 10.5815/ijigsp.2025.02.07.
Identification of Suspect Using Transformer and Cosine Similarity Model for Forensic Voice Comparison (2025):
This 2025 publication in *Security and Privacy (Wiley)* explores the application of **Transformer architectures** and **Cosine Similarity models** in suspect identification. Transformers, which excel at capturing long-range dependencies in data, are employed to extract complex speaker embeddings from audio signals. Once these embeddings are generated, a Cosine Similarity model is used to mathematically measure the likeness between the suspect’s voice and the evidentiary sample. This methodology provides a high-precision tool for forensic analysts, as the combination of self-attention mechanisms in Transformers and the distance-based comparison of Cosine Similarity enhances the model's ability to distinguish between similar-sounding voices. This work contributes to the modernization of digital forensics by moving beyond traditional speech analysis toward deep-learning-driven identification frameworks.
Link: https://doi.org/10.1002/spy2.70038.
Semi-Automatic Voice Comparison Approach Using Spiking Neural Network for Forensics (2025):
Featured in the *International Journal of Artificial Intelligence (IJ-AI)*, this peer-reviewed journal article provides an in-depth look at the **Spiking Neural Network (SNN)** approach mentioned in the author's patent. This research details the implementation of a semi-automatic system where AI handles complex feature extraction while allowing forensic experts to interpret the results. By focusing on SNNs, the study highlights a shift toward more neurologically plausible AI models that may offer superior performance in real-time forensic voice analysis. The paper underscores the system's ability to function as a decision-support tool, providing forensic practitioners with quantitative data to support their qualitative assessments of speaker identity in legal contexts. This Scopus-indexed work is a cornerstone of the author's research into efficient and accurate forensic identification technologies.
DOI: 10.11591/ijai.v14.i4.pp2689-2700
Semi-Automatic Approach through Diffusion Model Utilizing Siamese Neural Network for Forensic Voice Comparison (2026):
Published in *Franklin Open (Elsevier)*, this paper introduces a highly advanced approach using **Diffusion Models** and **Siamese Neural Networks**. Diffusion models, known for their powerful generative and denoising capabilities, are used here to enhance the quality of voice samples or to learn more precise latent representations. When paired with a Siamese Neural Network—a dual-branch architecture designed specifically for finding similarities between two inputs—the system becomes highly effective at "one-shot" or "few-shot" speaker verification. This is particularly useful in forensics where only one or two clear recordings of a suspect might exist. The study demonstrates how these modern AI techniques can significantly lower error rates in voice comparison, even when the input audio is contaminated by noise or recorded through different devices.
Link: https://www.sciencedirect.com/science/article/pii/S2773186326000435
Speech Processing and Analysis for Forensics and Cyber Crime: A Systematic Review (2023):
Published in *Cyber Crime in Social Media: Theory and Solutions (CRC Press)*, this chapter provides a comprehensive review of how **speech processing** is utilized in the fight against cybercrime. The authors systematically analyze existing methodologies for identifying suspects who use voice-based platforms for illicit activities. The review covers the evolution of forensic speech analysis from manual techniques to modern automated systems, highlighting the critical role of AI in processing large-scale social media data. By identifying gaps in current forensic practices, this work serves as a foundational roadmap for researchers and law enforcement agencies looking to improve their digital investigation capabilities through more advanced audio analysis tools.
Link: https://doi.org/10.1201/9781003304180.
Forensic Voice Comparison Approaches for Low Resource Languages (2024):
This chapter in *Automatic Speech Recognition and Translation for Low Resource Languages (Wiley)* addresses a critical gap in forensic science: the lack of robust tools for **low-resource languages**. Most forensic voice comparison systems are trained on major languages like English, making them less effective for regional dialects or minority languages. The authors propose specific approaches and adaptations to ensure that voice comparison algorithms remain accurate across diverse linguistic backgrounds. This research is vital for ensuring global justice, as it provides methodologies that can be applied in multi-linguistic societies where suspects may speak languages that are underrepresented in standard AI training datasets.
Link: https://doi.org/10.1002/9781394214624.ch9
Smart Home Automation and IoT Systems (2020):
The author’s early work includes two publications in 2020 focused on **Smart Home Automation Using Raspberry Assistant**. These papers explore the integration of IoT devices with central controllers like the Raspberry Pi to automate household tasks via voice or mobile interfaces. While these studies predate the author's primary focus on forensics, they established a strong foundation in audio input processing and real-time system responses. This early experience with voice assistants likely informed the author's later expertise in how voice commands and speaker characteristics can be captured and analyzed in digital environments.
Link: https://ijsrcseit.com/paper/CSEIT206365.pdf
A Survey on SMOTE-Deep: Novel Link-Based Classifier for Fraud Detection:
The study addresses a fundamental challenge in fraud detection: the issue of imbalanced datasets, where fraudulent instances are rare compared to legitimate ones. To mitigate this, the paper explores the SMOTE-Deep methodology, which integrates the Synthetic Minority Over-sampling Technique (SMOTE) with Deep Learning models. By generating synthetic examples of the minority (fraudulent) class, the system improves the training of a link-based classifier, allowing it to identify suspicious patterns and fraudulent links more accurately within large-scale data networks. This foundational work in handling data imbalance and developing robust classifiers provided the technical expertise necessary for the author's later, highly specialized advancements in forensic voice comparison and suspect identification.
Link: https://www.ijcsejournal.org/wp-content/uploads/2025/11/IJCSE-V5I1P1.pdf
Another early work (2014) focuses on Secure Attack Measure Selection and Intrusion Detection in Virtual Cloud Networks, in virtual cloud networks. These papers demonstrate a broad technical background in cybersecurity and network security, which complements the author's current forensic work. Understanding how data is secured and how fraudulent actors behave in digital networks provides a crucial context for forensic investigations into cybercrime and identity verification. Link: http://www.ijreat.org/Papers%202014/Issue8/IJREATV2I2073.pdf
The research paper titled Energy Efficient Cloud-Based Smart MOV (Mobile Social TV System): authored by Kruthika S.G. and Pradeep B.M., was published in the *International Journal of Engineering Research & Technology (IJERT)* as part of the NCITSF-14 Conference Proceedings at Jain University in May 2014. This early scholarly contribution explores the integration of mobile computing, social media, and cloud infrastructure to develop a "Smart MOV" platform. The study focuses on the critical challenge of maintaining **energy efficiency** in mobile devices while delivering high-bandwidth, interactive social TV services. To achieve this, the authors propose a **cloud-based architecture** that offloads heavy processing tasks from the mobile handset to cloud servers, thereby reducing power consumption and enhancing the user experience. This work is significant as it demonstrates the author's early research focus on optimizing distributed systems and mobile multimedia delivery. These foundational concepts in data processing and system efficiency provided a technical base for her subsequent transition into complex forensic voice comparison and AI-driven security algorithms. Link: https://www.ijert.org/energy-efficient-cloud-based-smart-mov-mobile-social-tv-system
Likelihood Ratio Based Voice Comparison Using Cepstral Coefficients and GAN (2023):
Presented at ICMLDE-2023, this research utilizes **Cepstral Coefficients**—a standard feature in speech processing—and **GANs** to calculate **Likelihood Ratios (LR)**. The Likelihood Ratio is a standard forensic metric used to express the strength of evidence in favor of one hypothesis over another (e.g., that the voice belongs to the suspect vs. a different person). By using GANs to model the variability of voice features, the research aims to provide more statistically sound evidence for courtroom use.
Link: https://doi.org/10.1016/j.procs.2024.04.287
The paper titled "Voice Comparison Approaches for Forensic Application: A review," presented at the ICSCCC-2023 IEEE Conference at Dr. Ambedkar National Institute of Technology, Jalandhar, serves as a comprehensive academic synthesis of the methodologies utilized in modern digital forensics. Authored by Kruthika S.G., Trisiladevi C. Nagavi, and P. Mahesha, this review evaluates a broad spectrum of speaker verification technologies, bridging the gap between traditional acoustic-phonetic analysis and emerging automated AI systems. doi:10.1109/ICSCCC58608.2023.10176553
The book chapter titled "Domain Specific Information Based Learning for Facial Image Forensics," authored by Chethana, T.C. Nagavi, P. Mahesha, and S.G. Kruthika, was published in 2024 in the Springer volume Adversarial Multimedia Forensics. This work represents a significant expansion of the research group's expertise, moving beyond audio-based analysis to address the critical challenges of facial image authentication.The research focuses on domain-specific information-based learning, a methodology designed to enhance the reliability of forensic tools in identifying manipulated or forged facial imagery. In the context of "Adversarial Multimedia Forensics," the study likely investigates how deep learning models can be trained to recognize subtle inconsistencies in facial data that standard algorithms might miss, particularly when faced with sophisticated adversarial attacks like deepfakes. By integrating specific domain knowledge into the learning process, the authors contribute a robust framework for maintaining digital integrity. This publication underscores the authors' versatile application of AI in the broader landscape of information security and forensic science. Link: https://doi.org/10.1007/978-3-031-49803-9_6.
Survey on Smart Home Automation Using Raspberry Assistance
Published in International Journal of Computer Science Engineering Technique, Vol. 5, Issue 2, and ISSN: 2455, 2020. It focus on the human race has always been in the quest for a better and easier life and ever since technology began heavily integrating itself into the society. From the appliances in our kitchen to the phones in our palms, there are many high tech devices that make our lives easier. But it would be more convenient if we could control these devices remotely. An automated system could control machines and perform tasks as per the convenience of the user from long distances. In this paper, we are presenting a proposed system for a Smart Home automation where the user can control home electronic appliances easily and with higher controllability using an android application. To design this system, we are using a Raspberry Pi module. The Raspberry Pi acts as a mini computer.
Link: https://www.ijeast.com/papers/496-501%2CTesma403%2CIJEAST.pdf
Enhancing Forensic Voice Comparison Utilizing MFCC and Deep Belief Network (2026):
This upcoming 2026 publication in *Procedia Computer Science* explores the synergy between **Mel-Frequency Cepstral Coefficients (MFCC)** and **Deep Belief Networks (DBN)**. MFCCs are highly effective at capturing the power spectrum of a voice, and when processed through a DBN, they allow the system to learn deep, non-linear relationships that are unique to an individual's vocal tract. This combination represents one of the author’s most advanced efforts to optimize the performance of forensic speaker verification systems.
Link: https://doi.org/10.1016/j.procs.2026.05.178
Developed a semi-automatic deep neural network framework for Forensic Voice Comparison (FVC) to improve speaker identification in forensic investigations.
Addressed key forensic challenges including background noise, channel mismatch, speaker variability, linguistic diversity, and voice disguise.
Proposed a GAN + MFCC + Logistic Regression framework for data augmentation and classification, achieving up to 85% accuracy.
Enhanced voice comparison performance by integrating MFCC, LPCC, GAN, and ANN, achieving up to 90% accuracy.
Designed a Spiking Neural Network (SNN)-based approach inspired by biological neural processing, achieving 94.21% accuracy.
Developed a Siamese Neural Network for similarity-based speaker verification using contrastive learning, achieving 96.02% accuracy.
Implemented a Transformer-based Wav2Vec 2.0 model for robust speech representation, achieving a similarity index of 0.9970–0.9995.
Evaluated Deep Belief Network (DBN) and Autoencoder models for feature learning and comparison on both .wav and .flac forensic datasets.
Performed comprehensive performance evaluation using accuracy, precision, recall, F1-score, and false positive rate.
Demonstrated that deep learning-based semi-automatic FVC significantly improves the accuracy, reliability, and robustness of forensic speaker identification.
Research Interests
Digital Signal Processing
Digital Forensics
Artificial Intelligence
Machine learning
Deep learning
Development of Voice Comparison Algorithms for Forensics:
Recipient of the prestigious DST WISE-KIRAN (Women Scientist) Ph.D. Fellowship from the Department of Science and Technology, Government of India, for conducting research on developing innovative voice comparison algorithms for forensic applications. TPN Number: 95844 | File No: DST/WISE-PhD/ET/2023/4(G) | Year: 2024
“Secure Attack Measure Selection and Intrusion Detection in Virtual Cloud Networks.”:
Received the Best Paper Presentation Award at the 3rd National Level Conference on Emerging Trends in Engineering Technology (ETET) held on February 21–22, 2014, at Jyothi Institute of Technology, Tataguni, Bangalore.
Served as an invited reviewer for multiple IEEE and international conferences and a Q1 journal, contributing expert peer reviews, ensuring research quality, and supporting excellence in global scientific publishing.
Served in diverse academic leadership roles, organized AICTE programmes, delivered invited technical talks, submitted competitive research proposals, earned postdoctoral recognition, and demonstrated strategic leadership through the RV University Program Director presentation.
Assistant Professor
School of Engineering
-
Mysuru