Saurabh Bhargava

Saurabh Bhargava is working with Shih-Chii on the SNF funded EARS project. He is focusing on source separation using a single microphone. He is currently evaluating both linear and nonlinear approaches. The linear approaches include eigenanalysis over the difference of covariance matrices of the training spectrograms of two speakers to obtain a set of basis. Source specific basis are then assigned based on the which source has higher variance in that subspace (basis). Another linear method proposed is the one which uses maximum likelihood method to demix the sources. He is also evaluating non-linear approaches including Non Negative Sparse Coding (NNSC).  All computation is performed using the Mocha-TIMIT database for two sources (one male and other female) and a single mixture.

Joachim Ott

My main research interest lies on machine learning, especially in the area of deep learning. Having a background in both biology and computer science, my fascination with deep networks results in optimizing and applying current algorithms, as well as getting inspiration from biological systems for new or improved versions.

Project
Currently, I am working on my master project thesis where I am developing new methods to optimize recurrent neural networks (RNNs) with limited bit precision for processing and classification of auditory data. This project is part of a larger project of our research group, and aims at allowing powerful spiking (recurrent) neural networks to run on low power devices.

Prof. Alejandro Linares-Barranco

Prof. Alejandro Linares-Barranco is a visiting professor from the University of Seville, Spain. He is working on programmable logic implementations of event-based algorithms.