Krishna Subramani

I am a Dual Degree Student at IIT Bombay majoring in Electrical Engineering and specializing in Signal Processing. I currently work in the Digital Audio Processing Lab supervised by Prof. Preeti Rao. My Master's Thesis is on Variational Parametric Models for Audio Synthesis.
I have had the opportunity to spend time as a visiting researcher at the Music Technology Group, Pompeu Fabra University Barcelona with Prof. Xavier Serra and the Cooperative Intelligence Lab (a joint research venture between Honda Research Institute Japan and Kyoto University) with Prof. Hideaki Shimazaki.

Email  /  CV  /  LinkedIn  /  GitHub

Krishna Face


Energy-Weighted Multi-Band Novelty Functions for Onset Detection in Piano Music
Krishna Subramani, Srivatsan Sridhar, Rohit M A, Preeti Rao
National Conference on Communications 2018
Paper / DOI

Proposed the use of energy-based weighting of multi-band onset detection functions and the use of a new criterion for adapting the final peak-picking threshold to improve detection of soft onsets in the vicinity of loud notes. Also proposed a grouping algorithm to reduce spurious onsets.

We presented our work at the conference as an oral presentation

Generative Audio Synthesis with a Parametric Model
Krishna Subramani, Alexandre D'Hooge, Preeti Rao
ISMIR 2019 Late Breaking/Demo
Abstract / arXiv / Audio Examples

Proposed a parametric representation for audio corresponding more directly to its musical attributes such as pitch, dynamics and timbre. For more control over generation, we also proposed the use of a conditional variational autoencoder which conditions the timbre on pitch.

We presented our work as a poster in the ISMIR 2019 Late Breaking/Demo session

Submitted / In Preparation

VaPar SYNTH - A Variational Parametric Model for Audio Synthesis
Krishna Subramani, Preeti Rao Alexandre D'Hooge
Submitted to ICASSP 2020, Under Review
Paper / Audio Examples

We presented VaPar Synth - a Variational Parametric Synthesizer which utilized a conditional variational autoencoder trained on a suitable parametric representation.

Learning Complex Representations from Spatial Phase Statistics of Natural Scenes
HaDi Maboudi, Krishna Subramani, Hamid Soltanian-Zadeh, Shun-ichi Amari, Hideaki Shimazaki
To be submitted to Neural Networks

We introduced a generative model for phase in visual systems, and propose a complex domain based maximum likelihood estimation procedure for parameter estimation. We derive analytical gradient expressions for maximum likelihood estimation using Wirtinger Calculus (detailed in our appendix)

I presented the initial part of this work as an oral presentation at Honda Research Institute

Research Experience
Variational Parametric Models for Audio Synthesis
Report / Repository / Review of Generative Models for Audio Synthesis

(Ongoing)My Master's thesis. Working on generative parametric models for audio synthesis.

I presented the work done by me so far as an oral presentation for the first stage defense of my Master's thesis

Compression using Graph Signal Processing
Presentation / Report / Notebook Demonstration of the Code

Inspired by the authors work on compressing distributed data by modeling it as graphs, we decided to try out and extend the idea to a new dataset by performing outlier removal, data imputation and clipping, and were able to successfully compress it at a significantly faster rate(~8x speedup) than conventional techniques.
Data Sonification using Granular Synthesis
Report / Code Repository

Implemented a real-time granular synthesizer on Pure Data. Also ideated a scheme where the parameters of the synth could be controlled externally with data, thus 'sonifying' the data!

Kuramoto Model and Oscillatory Networks
Image Recall

Implemented an image retrieval(memory recall) task using the Kuramoto Model. Extended this to try out graph coloring using Oscillatory Networks. Presented a poster on the same.

The Master Yoda to us Padawans.