Shreeram

Shreeram Chandra

Electrical Engineering

graduate student @UTD

Bio.

I am a second year Ph.D student advised by Prof. Berrak Sisman at the Speech and Machine Learning Lab, UT- Dallas. My research revolves around building generative deep learning models that improve controllability in expressive/emotional speech synthesis. I am interested in label and data scarce scenarios and I tackle multi-modal speech-language problems in my work.

Quick Links.

Education

Fun stuff

Skills and Coursework

Education

July 2017 - May 2021

Aug 2022 - May 2027

Electronics and Communication Engineering

Computer Engineering

(Signals and Systems track)

CGPA : 8.42/10

GPA : 3.94/4.0

Publications

Chandra, S. S., Du, Z., & Sisman, B. (2024). Exploring speech style spaces with language models: Emotional TTS without emotion labels. arXiv preprint arXiv:2405.11413.

Salman, A. N., Du, Z., Chandra, S. S., Ulgen, I. R., Busso, C., & Sisman, B. (2024). Towards Naturalistic Voice Conversion: NaturalVoices Dataset with an Automatic Processing Pipeline. arXiv preprint arXiv:2406.04494.

S.S Chandra, S.Arthi and T.V Sreenivas, ”Differences in the spatial integration of discrete/continuous spectral signals,” Proc. International Conference on Electronics, Computing and Communication Technologies (CONECCT), Bengaluru, India, Jul. 2021. (Paper) (Video) (Slides)

S. S. Chandra, A. Upadhye, P. Saravanan and S. Gurugopinath, ”Deep Neural Network Architectures for Spectrum Sensing Using Signal Processing Features,” Proc. IEEE International Conference on Distributed Computing, VLSI, Electrical Circuits and Robotics (DISCOVER), Nitte, India, Nov. 2021. (Paper) (Slides)

P. Saravanan, S. S. Chandra, A. Upadhye and S. Gurugopinath, ”A Supervised Learning Approach for Differential Entropy Feature-based Spectrum Sensing,” Proc. International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), Chennai, India, Mar. 2021. (Paper) (Video) (Slides)

A. R. Bharadwaj, S. S. Chandra, D. S. Nair, A. R. Hatim and A. Ravikumar, “Automated mythological scene recognition using machine learning and graphs ”, 2020 International Conference on Artificial Intelligence and Signal Processing (AISP), Amaravati, India, 2020, pp. 1-5, Jan 2020. (Paper) (Slides)

A. Upadhye, P. Saravanan, S. S. Chandra and S. Gurugopinath, ”A Survey on Machine Learning Algorithms for Applications in Cognitive Radio Networks,” Proc. International Conference on Electronics, Computing and Communication Technologies (CONECCT), Bengaluru, India, Jul. 2021. (Paper)(8-page version)(Slides)

Experience

INTEL July 2021 - July 2022

Graphics Hardware Engineer

Working in the pre-silicon validation team to build a software modeling simulator to mimic the hardware behavior in the GPU.

Worked on the pixel processing pipeline, where I learnt about the various image filters that need to be applied to render a visually appealing image on the display panel.

Indian Institute of Science, Speech and Audio group Jan 2021 - July 2021

research intern

Worked under Prof. T.V Sreenivas in the Speech and audio group, Department of Electrical Communication, IISc.

Generated audio signals using Digital Signal Processing algorithms and designed listening experiments to understand signal properties affecting perceived source width.

Used source filter model to generate synthetic vowels after extracting filter parameters using LPC analysis on recorded natural vowels.

Designed algorithms to calculate the noise floor difference (NFD) between noisy and noiseless speech signals. (Github)

Zero crossing instantaneous frequency estimation to explain the source width widening of shaped noisy vowels by decomposing the vowel into multiple band pass signals.

Performed loudspeaker compensation by designing inverse filters using LP analysis.

Also built a Matlab GUI to perform multi-channel loudness equalisation.

INTEL may 2020 - nov 2020

Undergraduate intern

Wrote a parser in C++ to reproduce bugs caught on test benches from the customer.

Received recognition by manager and team for quality code and quick turn-around time.

Worked in the Display GPU modelling team.

Generated code coverage data analysis using python scripts.

Worked on a test application that was able to interface various components of the

Display architecture.

MICROSOFT INNOVATION LAB, PESU may 2019 - aug 2019

Research project intern

Worked on project Mythodex, a context-based mythological scene mapping model.

Learnt about neural networks, graphs and object detection techniques like YOLO,

Single Shot Detector, Faster R-CNN etc.

Your Section Header Goes Here

Your Sub-Header Goes Here

This is prime space! Use it to elaborate on your attention-grabbing section title. Explain what this section is about, share some details, and give just the right amount of information to get the audience hooked.

Your Sub-Header Goes Here

Add a Call-to-Action

Fog, Road, Highway, Tar, Central Reservation, Landscape

Projects

ML Algorithms for Cognitive Radio networks

An research project that tackled issues like spectrum sensing, spectrum allocation

and spectrum scheduling in CR networks using a machine learning framework.

Read further

Code

Audio Analysis on Carnatic Music

Calculation of similarity index based on features specific to Carnatic music characteristics like tonic, pitch histograms, pattern of note occurrence, gamakas. Developed an algorithm to identify the tonic of the song without any meta-data.

Code

Read further

Audio perception experiments

We explore the auditory perceptual property of virtual source width (VSW) expansion of synthetic vowels /a/, /i/ and /u/ and compare it with that of natural vowels. We also modify the synthetic vowels by adding white noise at the excitation or at the output. We observe that continuous vowel spectra creates a more stable VSW perception than discrete spectra.

Code

Read further

Projects

Mythological scene mapping

The aim of the project was to automate the identification of scenes from Indian mythology in works of art. Images passed through neural networks undergo multiple levels of filtering and result in a list of elements from the image that bear meaningful relationships.

The end result of the project is an accurate identification and description of the scene being depicted.

Code

Rail fence cipher

A rail fence cipher algorithm was implemented from scratch in python.

Code

Lossless image compression algorithms

The following compression and decompression algorithms were implemented from

scratch on images: 1) Huffman encoding 2) Run-length encoding 3) Constant Area

encoding 4) LZW encoding

Code

Research Interests