I'm a Masters student in Electrical Engineering specializing in signal Processing emphasizing in the areas of Speech/Audio Processing, Computer Vision, Machine Learning, Deep Learning. I graduated in May 2020 and I'm currently in active pursuit of full time opportunities.I have taken several courses related to these domains like Speech and Audio Processing and Perception, Physics Based Computer Vision, Deep Learning Media Processing & Understanding, Statistical Machine Learning, Digital Image and Video Processing during my time at Masters.
I have worked on several state of the art Research projects in these domains as well.
-
Experience
Teaching Assistant:
School of Mathematical and Statistical Sciences, Arizona State University, Arizona, USA:
• Guided and managed a cohort of 125 students towards academic excellence. Assisted professor in evaluating performance of students by grading their work.
Research Intern:
Research Centre Imarat (Defense Research Development Organization)-Hyderabad, India
• Design of ‘UART protocol using Simulink’ and implementation using FPGAs.
• Design and Fabrication of ‘Metamaterial Absorbers’ using Split Ring Resonators. Implemented using CST Studio Suite
Efftronics-Vijayawada, India
• Design and Maintenance of Data loggers.
-
Projects
Deep HDR Video Reconstruction using a Single Exposure
• Devised a convolutional neural network (CNN) design in the form of a Hybrid Dynamic Range Autoencoder (VGG16) in python using Tensor Flow, Tensor Layer to reconstruct HDR video from a given LDR video of single exposure. Pre-trained model using Transfer learning.
• Data Augmentation is processed in C++ using OpenCV and a realistic camera curve called Virtual Camera. MIT places dataset is used.
• State of the Art results attained by using a loss function based on Illuminance and Reflectance components of every frame and Alpha-Blending.
Transfer Learning to detect Tuberculosis and Lung Cancer
• Accomplished using PyTorch and Tensor Flow libraries on the CheXNet dataset containing over 100,000 frontal Chest X-rays obtained by Stanford students to improve accuracy of Pretrained model using a 121-Layer Convolution neural network.
• Streamlined the model towards image classification by pre-training on following datasets: ResNet, ImageNet, DenseNet
• The Accuracies attained to detect Tuberculosis was 91% and to detect Lung Cancer was 46%.
Speech Emotion Recognition
• Constructed a Neural Network model in Python to classify various emotions of a given speech segment by extracting several speech features like MFCCs, Spectral flatness, Spectral Contrast, Zero-Crossing Rate, Energy. These models are trained on Feed Forward Multi-Layer Perceptron, 11-layer Convoluted Neural Networks, Support Vector Machines, Random Forest Classifier for comparative analysis.
• Datasets used for training were RAVDNESS and EMO-DB. Accuracy of 92% was obtained during classification.