Shrishti Saha Shetu

- Online Speaker Independent Audio-Visual Speech Enhancement - Real Time Deep Noise Suppression - Evaluation of Different Objective Measures

Title: An Empirical Study of Visual Features for DNN based Audio-Visual Speech Enhancement - Development of audio-visual data pipeline from various open-source databases - Development of the Google speech separation model for our AV speech enhancement system - Training with different model architectures and visual features for various datasets - Objective evaluation and comparison with different objective measures

Title: Listening to the Cocktail Party Effect: Decoding Auditory Attention using EEG Signal. • Feature extraction from EEG data and Speech signal. • Creating deep learning networks for specified auditory attention decoding problem using EEG and Speech features. • Investigating different input data structures and network topologies using Keras framework.

Title. Blind Reverberation Time Estimation (RT60) in Frequency Subbands. - Reviewing existing literature and implementing different existing models for comparison - Creating a new non-blind algorithm for automatic T60 estimation using the Schroeder method - Development of DNN based system for subband reverberation time estimation - Reverse Estimation of fullband T60 from Subband Measures

• Preparing Jupyter notebook codes for different music processing topics. • Creating a new algorithm for segmentation of sheet music images for automatic score follower.

4G / 5G Mobile Communications, Auditory models, Digital Signal Processing, Image and video compression, Mobile communications, Music signal processing - analysis and synthesis, Pattern recognition, Signal analysis , Speech and Audio Signal Processing and Statistical signal processing