PhD in Signal Processing (Speech, Image)

Integrated B.Tech + M.Tech in Electronics and Communication Engineering

Doctoral Candidate (Marie Sklodowska-Curie Fellow)

Experience
• Ideated and implemented an innovative algorithm for segmenting lip and hand movements from Cued Speech videos. Published and presented the results of this work at INTERSPEECH 2023
• Engineered end-to-end ACSR systems, achieving a 15.2% improvement in recognition accuracy at phonetic level and presented the findings at ICASSP 2022
• Currently exploring generative models for small data to develop the first-ever CS Generation framework
• Released a curated clean mono-blications
cuer dataset for CS (DOI: 10.5281/zenodo.8392608)
• Actively involved in data collection efforts to establish the first French multi-cuer corpus

Publlications
[1] S. Sankar, D. Beautemps, F. Elisei, O. Perrotin, T. Hueber, “Investigating the dynamics of hand and lips in
French Cued Speech using attention mechanisms and CTC-based decoding”, INTERSPEECH 2023, DOI: 10.1109/ICASSP43922.2022.9746976
[2] S. Sankar, D. Beautemps, T. Hueber, “Multistream Neural Architectures for Cued Speech Recognition using
Pre-trained Feature Extractor and Constrained CTC Decoding”, ICASSP 2022, DOI: 10.21437/Interspeech.2023-1669
[3] Public access dataset released with [2] on Zenodo – DOI: 10.5281/zenodo.8392608

Secondments
• Collaborated with Ivès to test and implement an online data collection software for CS and sign language
• Visiting researcher at the University de Libre Bruxelles – EU collab project for intersectoral exposure among researchers

Project Associate

• Devised and deployed a highly effective end-to-end Indian-English ASR (https://www.iitm.ac.in/speech/asr_new/) system for the automatic transcription of NPTEL (https://nptel.ac.in/) courses, significantly improving transcription accuracy and streamlining the learning experience
• Implemented the baseline for the Hindi ASR Challenge (https://sites.google.com/view/asr-challenge) and monitored the submissions
• Worked on enhancing speaker diarization systems and extracting speaker embeddings for voice conversion

Research Masters Student (Thesis)

• Revamped adaptive filtering algorithms with kernel methods to reduce non-linear echo return loss in audio devices
• The above project culminated as a journal publication in Applied Acoustics (DOI: 10.1016/j.apacoust.2020.107329)
• Pioneered the development of a novel collaborative scheme for non-linear stereophonic acoustic echo cancellation, a key contribution acknowledged in the Journal of AIHC (DOI: 10.1007/s12652-021-03647-2)

Research Intern

• Developed a compact module of a Grapheme-to-Phoneme converter and interfaced it with the VocalTractLab (https://www.vocaltractlab.de/)
• Ported SEQUITUR algorithm from Windows and revamped for a Linux platform

Best Project Award

Best project award of the graduating class of 2020, Electronics Department, IIITDM

Honors Student

Consistent Hinors student with 9+/10 CGPA in every semester

GFF Scholarship

Among 10 awarded annually by TU Dresden for international intern students

Summer Research Fellowship

Awarded by Indian Academy of Sciences Among the 150 students to receive the fellowship nationwide

Marie Sklodowska-Curie Fellowship