Research Associate, Machine Learning for Acoustic Frontend
Fraunhofer IIS
Graduate Studies: Master's in Communication and Multimedia Engineering (FAU Erlangen Nuremberg, Germany, class of 2020)
Undergraduate Studies: B.Tech in Electrical and Electronics Engineering (National Institute of Technology, Jamshedpur, India, class of 2017)
Areas of interest entail: Machine Learning, Multimedia Signal Processing, Computational Neuroscience, AI for Health and Biomedical Research.
Papers
This section covers all my published conference and journal papers.
Projects
This section covers all the projects I have completed and some demos of those projects.
Deep Learning Based Noise Suppression
In this project we present a novel deep learning method to suppress any kind of noises present in a speech signal. The Deep learning model used for this project use a covulational recurrent neural network and the model was trained with 500 hours data provided by Microsoft DNS challenge
Visit the project page for more details and listen to the enhanced audio samples.An Empirical Study of Visual Features for Deep Learning based Audio-Visual Speech Enhancement
This project was the part of my master thesis work. In this project we reimplemented the AV speech separation model presented in Google Looking to Listen Project . Furthermore, we also proposed a raw lip images based AV speech enhancement model and compared the results with Google model. This project also covers a study of other visual features, such as, lip embedding and also the the effect of visual features on speech denoising tasks.
Visit the project page for more details and listen to the enhanced audio-visual samples.Blind Estimation of the Subband Reverberation Time
The research work for this project was done at Multimedia Communication and Signal Processing department at FAU Erlangen as a part of my research internship. In this project, we studied many novel approaches for subband T60s estimation. We also proposed two diffrent methods for blind subband and fullband T60 estimation. Please visit the projcet page for more details.
Lanes Detection for Autonomus Driving
The difficulty with the road-lane markings is that there is no labeled dataset of them and creating such dataset would cost millions of dollars. In this project, we solved this problem by creating a dataset of simulated images and then intermixed with a dataset of real images that contain no road.
Acoustic Echo Cancellation with Double-Talk Detection
The need for acoustic echo cancellation arises whenever a loud speaker and a microphone are placed together in nearby vicinity of each other and as a result the electroacoustic circuit may become unstable and produce undesirable howling. So to solve this problem, in this project we use the regularized Normalized LeastMean Square(NLMS) algorithmto generate the coefficients of the adaptive filter and to have a control over the filter adaption we couple it with a double talk detector usingMagnitude Squared Coherence(MSC) to extract the local speech signal.
Overfitting and Regularization for Wireless Resource Management.
In this work, we address both the theoretical and practical aspects of overfitting and regularization with applications to wireless resource management. We first discover the theoretical reasoning of overfitting and how regularization helps and then on implementation side we show the effects of regularization on a practical DNN-based approximatin
My Skills: Experience and Learning
This section highlights my courses and experience and gives a general overview on the skills I have gathered over the recent past.
Testimonials and Awards
This section has excerpts and snapshots of recommendations, testimonials and awards from people I have interacted with in my academic and professional journey.
Prof. Dr. ir. Emanuël Habets
Associate Professor for Perception-based Spatial Audio Signal Processing
He stood out among his peers for his egarness to engage in the process of learning and discovery.
Prof. Dr. Niranjan Kumar
Professor, EEE, NIT, Jamshedpur.
He distinguished himself as one of the best students in the class and laboratory, with a keen grasp of basic facts coupled with
mathematical knowledge, which he applies efficiently to problem solving.
Dr. Madhu Singh
Associate Professor, EEE, NIT, Jamshedpur.
What am I currently learning and doing?
Projects
- Realtime speech denoising for streaming systems, IoT devices and hearing aids
- Online audio-visual speech enhancement
- Personal Project:Improving the implementation of Deep-Channel uses deep neural networks to detect single-molecule events from patch-clamp data