About Me

I am currently working as a Machine Learning Scientist at Twilio Inc. focussing on the development of speech recognition models in scale, and leading the development effort for the manual data asset acquisition pipeline within our infrastructure. I worked as a Speech Scientist at Capio Inc. focussing on implementation of Auto-ML techniques for speech recognition models. I was a PhD. candidate at Carnegie Mellon University in the Department of Electrical and Computer Engineering, based off the Silicon Valley Campus. I graduated with an MS in Electrical and Computer Engineering from Carnegie Mellon University in 2011 and a B.E. in Electronics and Telecommunications in 2010 from the University of Pune, India. My past industry experiences includes an internships at Baidu Silicon Valley Artificial Intelligence Lab, and Lenovo Speech Labs. My research is focussed on the development and analysis of techniques that can be used to speed up hyper-parameter optimization, specifically for deep neural networks for tasks in speech recognition and image processing on various computing architectures.

Contact Details

Akshay Chandrashekaran
Carnegie Mellon University
Bldg 19, Nasa Ames Research Center
Room 1035
Moffett Field, CA 94035 USA
Email: web@akshayc.com
Group Website: Speech@SV

Education

Carnegie Mellon University

Ph.D. in Electrical and Computer Engineering January 2012 - Present

I am a Ph.D. Candidate in ECE. My research is focussed on hyper-parameter optimization strategies to build deep neural network (DNN) systems for speech recognition on embedded platforms. I am currently working on methods to utilize learning curves from previously completed hyper-parameter configurations to predict the performance for a new hyper-parameter configuration.

Carnegie Mellon University

M.S. in Electrical and Computer Engineering August 2010 - December 2011

I have completed my M.S. in Electrical and Computer Engineering with focus on Signal Processing and Machine Learning

Vishwakarma Institute of Technology

B.E. in Electronics and Telecommunication Engineering August 2006 - May 2010

I have completed my B.E. in Electronics and Telecommunication Engineering with focus on Signal Processing and Embedded Systems.

Work

Capio Inc.

Speech Scientist June 2017 - Present

I am working as a speech scientist on the following:
Acceleration of optimization of neural network-based speech recognition models.
Rapid development of Speech recognition models for multiple low-resource languages.
Integration of Automated hyper-parameter optimization techniques for ASR model combination.
Exploratory work on online speaker change detection.
Integration of dynamic word addition to online speech recognition.
Integration of automated hyper-parameter optimization for decoder hyper-parameters.

Baidu SVAIL

Research Intern May 2016 - August 2016

During the course of my internship, I worked on importance sampling-based data sampling techniques to improve training time for speech recognition.
I also worked on improving the in-house speech recognition toolkit for better visualization of results.
I implemented a framework to allow different data selection strategies for deep learning.

Lenovo

Research Intern October 2013 - May 2014

As a research intern at Lenovo Speech Labs, I worked on the development of an Android SDK that could allow developers to enable multi-modal interaction (speech and touch) within their applications.
I am a co-inventor in three patents that have resulted from this work.

Patents

Publications

  1. A. Chandrashekaran, I. Lane, “Auto-ML for Feature, Acoustic Model and Decoder Parameter Optimization”, GTC 2018 Poster.
  2. K. Han, A. Chandrashekaran, J. Kim, I. Lane, “Densely Connected Networks for Conversational Speech Recognition”, Interspeech 2018 (submitted).
  3. K. Han, A. Chandrashekaran, J. Kim. I. Lane, “The CAPIO 2017 Conversational Speech Recognition System”, arXiv preprint 1801.00059.
    [ Paper]
  4. A. Chandrashekaran, I. Lane, “Speeding up Hyper-parameter Optimization by Extrapolation of Learning Curves using Previous Builds”, ECML 2017.
    [ Paper] [ Code ]
  5. A. Chandrashekaran, I. Lane, “Hierarchical Constrained Bayesian Optimization for Feature, Acoustic Model and Decoder Parameter Optimization”, Interspeech 2017.
    [Paper]
  6. A. Chandrashekaran, I. Lane, “Automated optimization of decoder hyper-parameters for online LVCSR”, Spoken Language Technologies Workshop (SLT 2016).
    [ Paper ] [ Poster]
  7. A. Chandrashekaran, I. Lane, “Automated Feature and Model Optimization for Task-specific Acoustic Models”, BayLearn 2015 (Poster).
    [Poster Abstract]
  8. D. Cohen, A. Chandrashekaran, I. Lane, & A. Raux, "The HRI-CMU corpus of situated in-car interactions." In Situated Dialog in Speech-Based Human-Computer Interaction (2016).
    [Paper]
  9. I. Lane, V. Prasad, G. Sinha, A. Umuhoza, S. Luo, A. Chandrashekaran, A. Raux, “HRItk: the human-robot interaction ToolKit rapid development of speech-centric interactive systems in ROS.” NAACL-HLT Workshop on Future Directions and Needs in the Spoken Dialog Community: Tools and Data (NAACL-HLT 2012). Association for Computational Linguistics.
    [Paper]

Skills

  • C
  • C++
  • Python
  • Opencl
  • Cuda
  • OpenMP
  • Java
  • Matlab
  • C#