Deepak Baby

Deepak Baby

Applied Scientist

Amazon Alexa

Biography

Deepak Baby is with the Amazon Alexa team as an Applied Scientist since October 2020.

He was previously working as a post-doctoral researcher with Prof. Hervé Bourlard at Idiap Research Institute between May 2019 and June 2020. Prior to that he was a post-doctoral researcher with Prof. Sarah Verhulst at WAVES, Department of Information Technology, Ghent University between February 2017 and April 2019, where he was working on DNN-based hearing restoration strategies for hearing impairment.

Deepak Baby received his PhD in Electrical Engineering (Thesis title : Non-negative Sparse Representations for Speech Enhancement and Recognition) from ESAT/PSI group in KU Leuven in November 2016 under the supervision of Prof. Hugo Van hamme. Deepak received the Masters degree in Communication Engineering from Dept. of Electrical Engineering, Indian Institute of Technology Bombay in 2012 and Bachelors degree in Electronics and Communication Engineering from College of Engineering Trivandrum, India in 2009.

Interests

  • Speech Recognition
  • Speech Enhancement
  • Deep Learning
  • Signal Processing

Education

  • PhD in Electrical Engineering, 2016

    KU Leuven, Belgium

  • MTech in Communication Engineering, 2012

    IIT Bombay, India

  • BTech in Electronics and Communication Engineering, 2009

    College of Engineering Trivandrum, India

Experience

 
 
 
 
 

Applied Scientist

Amazon

Oct 2020 – Present Aachen, Germany
Applied Scientist in the Alexa Speech team.
 
 
 
 
 

Post-doctoral Researcher

Idiap Research Institute

May 2019 – Jun 2020 Martigny, Switzerland
Worked with Prof. Herve Bourlard on sparse and hierarchical representations for speech modeling and automatic speech recognition.
 
 
 
 
 

Post-doctoral Researcher

Ghent University, Belgium

Feb 2017 – Apr 2019 Ghent, Belgium

Working on the ERC project RobSpear developing deep learning aided strategies for correcting non-linear hearing deficits. Responsibilities include:

  • Designing experiments
  • Developing deep learning-based architectures
  • Supervising PhD students and masters thesis
 
 
 
 
 

Visiting Researcher

Nuance Communications, Inc.

Apr 2015 – Jun 2015 Merelbeke, Belgium
Investigated the previously proposed exemplar-based speech enhancement approaches as front-end for Nuance’s ASR tasks on automotive data.
 
 
 
 
 

Visiting Researcher

Tampere Univsersity of Technology

Jun 2013 – Aug 2013 Tampere, Finland
Investigated the use of Modulation Envelope features for feature enhancement to improve the noise robustness of Automatic Speech Recognition systems with Audio Research Group, Dept. of Signal Processing, TUT. Proposed an approach using coupled dictionaries for exemplar-based speech/feature enhancement.

Invited Talks & Lectures

End-to-end speech enhancement models using deep learning

Jan 10, 2019 12:00 AM

Data driven techniques for single-channel speech enhancement

Mar 26, 2018 12:00 AM

An introduction to deep learning for beginners

Jan 7, 2017 12:00 AM

Exemplar-based Speech Enhancement and its Applications to ASR

Jul 2, 2015 12:00 AM

Noise-robust ASR with deep learning: History, principles and new results

Jul 2, 2015 12:00 AM

Academic Service

Mentoring

2018 – 2019 : Arthur Van Den Broucke, MS Student, Ghent University, Belgium
Masters Thesis : Deep learning techniques for individualized auditory profiles

2014 – 2015 : Joris Verhaegen, MS Student, KU Leuven, Belgium
Masters Thesis: NMF-based reduction of background sounds in TV shows for better ASR

Summer 2015 : Shubham Sharma, Summer Intern, KU Leuven, Belgium
Topic: De-reverberation using non-negative matrix factorisation

Summer 2012 : Rainu V. Philip, Anjaly Elias, Summer Interns, IIT Bombay, India
Topic: Face detection using sparse representations based on compressive sensing algorithms

Reviewing

Journals

  • IEEE Transactions on Audio, Speech and Language Processing
  • IEEE Transactions on Signal Processing
  • IEEE Signal Processing Letters
  • Computer, Speech & Language
  • Speech Communication
  • Journal of Acoustical Society of America
  • Neural Networks (Elsevier)

Conferences

  • INTERSPEECH
  • IEEE-ICASSP
  • IEEE International Conference on Multimedia & Expo (ICME)

Recent Posts

NICE- Non-linear Independent Components Estimation: Insights and Implementation in Keras

Keras implementation can be found here. Flow-based deep generative models have not gained much attention in the research community when compared to GANs or VAEs. This post discusses a flow-based model called NICE, its advantages over the other generative models and finally an implementation in Keras.
Last updated on Oct 15, 2020 8 min read

Implementing Variational Autoencoders: Insights and some tricks

This post is a summary of some of the main hurdles I encountered in implementing a VAE on a custom dataset and the tricks I used to solve them. The keras code snippets are also provided.
Last updated on Oct 15, 2020 5 min read

Understanding Variational Autoencoders and Implementation in Keras

Variational Autoencoders (VAEs)[Kingma, et.al (2013)] let us design complex generative models of data that can be trained on large datasets. This post is about understanding the VAE concepts, its loss functions and how we can implement it in keras.
Last updated on Oct 15, 2020 10 min read
See all posts

Projects

CoNNear

A convolutional neural-network model of human cochlear mechanics and filter tuning for real-time applications

NICE Keras

Implementation of NICE: Nonlinear Independent Components Estimation in Keras

iSEGAN

Tricks to improve SEGAN performance. Eveything is re-implemented into Keras with Tensorflow backend.

SERGAN: Speech enhancement relativistic generative adversarial network

A fully convolutional end-to-end speech enhancement system with GANs

Joint dereverberation and denoising using NMD

MATLAB impelementation of a joint dereverberation and denoising algorithm based on non-negative matrix deconvolution

CHiME-3 CNN ASR

Kaldi script for CNN-DNN evaluation on CHiME-3 data

Publications

Google Scholar Profile

Fotios Drakopoulose, Deepak Baby, Sarah Verhulst (2021). A convolutional neural-network framework for modelling auditory sensory cells and synapses. Nat Commun Biol.

Preprint PDF Code DOI

Deepak Baby, Herve Bourlard (2021). Speech Dereverberation using Variational Autoencoders. IEEE-ICASSP.

Deepak Baby, Arthur Van Den Broucke, Sarah Verhulst (2021). A convolutional neural-network model of human cochlear mechanics and filter tuning for real-time applications. Nat Mach Intell.

Preprint PDF Code DOI

Arthur Van Den Broucke, Deepak Baby, Sarah Verhulst (2020). Hearing-Impaired Bio-Inspired Cochlear Models for Real-Time Auditory Applications. INTERSPEECH.

PDF DOI

See all publications

Contact