Apoorv Vyas

Hello! I am Apoorv Vyas, currently a Ph.D. student at EPFL, and research assistant in Speech and Audio Processing at Idiap Research Institute under the joint supervision of Prof. Hervé Bourlard and Prof. François Fleuret.

Prior to joining Idiap, I received my Bachelor's in Electrical Engineering from Indian Institute of Technology Guwahati. After which I joined Intel Labs (Bangalore) where I worked on Compressed Sensing for power-efficient communications in body sensor networks and wireless sensor networks. I also worked on detecting out-of-distribution input to a deep neural network.

Email  /  CV  /  Google Scholar  /  Github  /  Twitter  /  LinkedIn

profile photo
Research

I am interested in machine learning, speech recognition, and computer vision.

The goal of my Ph.D is to improve the performance of end-to-end automatic speech recognition (ASR) models with a special focus on the low to medium resource datasets.

Publications
Comparing CTC and LFMMI for Out-of-Domain Adaptation of Wav2vec 2.0 Acoustic Model
Interspeech , 2021
paper

Comparing sequence discriminative criterion for adaptation of wav2vec 2.0 model.

Lattice-Free MMI Adaptation of Self-Supervised Pretrained Acoustic Models
ICASSP, 2021
paper / code / bibtex

Using MMI loss to adapt pre-trained acoustic models to low resource datasets.

Fast Transformers with Clustered Attention
NeurIPS, 2020
paper / poster / blog / colab / code / bibtex

Scaling Attention to long sequences by clustering queries.

Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention
ICML, 2020
paper / video / slides / colab / code / bibtex

Scaling Attention to long sequences with kernelized linear attention.

Pkwrap: a PyTorch Package for LF-MMI Training of Acoustic Models
S. Madikeri, S. Tong, J. GOMEZ, A. Vyas, P. Motlic, H. Bourlard,
arXiv, 2020
paper / code / bibtex

PyTorch package to expose Kaldi functionalities and LF-MMI loss

Unbiased Semi-supervised LF-MMI Training Using Dropout
S. Tong, A. Vyas, P. Garner, H. Bourlard,
Interspeech, 2019
paper / poster / bibtex

Semisupervised Training by combining multiple hypotheses with Dropout.

Analyzing Uncertainties in Speech Recognition Using Dropout
A. Vyas, P. Dighe, S. Tong, H. Bourlard,
ICASSP, 2019
paper / poster / bibtex

Unsupervised word error rate estimation by analyzing multiple hypotheses with Dropout.

Out-of-Distribution Detection Using an Ensemble of Self Supervised Leave-out Classifiers
A. Vyas, N. Jammalamadaka, X. Zhu, D. Das, B. Kaul, T. Willke
ECCV, 2018
paper / poster / bibtex

Detecting out-of-distribution input by entropy maximization.

Power Efficient Compressive Sensing for Continuous Monitoring of ECG and PPG in a Wearable System
V. Natarajan, A. Vyas,
WFIOT, 2016
paper / bibtex

Using compressive sensing for energy efficient signal acquisition and denoising on wearable devices.

Commercial Block Detection in Broadcast News Videos
A. Vyas, R. Kannao, V. Bhargava, P. Guha
ICVGIP, 2014
paper / bibtex

Detecting commercials in TV News using hand-crafted features and SVM.


Thanks to Jon Barron for the template.