CV - Saurabh Kumar

Saurabh Kumar — Download this CV as a PDF
Google Scholar | GitHub | LinkedIn | ORCID
Email: saurabhk0317@gmail.com

Education

M.Tech. (Research) in Electrical Engineering
Indian Institute of Science (IISc), Bengaluru, India
2026–Present

Advisor: Prof. Prasanta Kumar Ghosh (SPIRE Lab)
GATE 2026 (Data Science and AI): AIR 362 / 69,242 (top 0.5%)
Proposed a CTC training criterion with token-level wildcard arcs that tolerates localized transcription ambiguity (pronunciation, spelling, lexical variants), achieving a 9.45% average relative WER reduction over CTC across multiple languages and corpora (preprint).

B.Tech. in Electronics and Communication Engineering
National Institute of Technology (NIT), Patna, India
2016–2020

Research Associate
SPIRE Lab, Indian Institute of Science (IISc), Bengaluru, India
July 2021 – July 2026
Advisor: Prof. Prasanta Kumar Ghosh

RESPIN: Speech Recognition in Agriculture and Finance for the Poor in India (funded by the Gates Foundation)

First author of the RESPIN-S1.0 database paper published at NeurIPS 2025 (Datasets and Benchmarks Track); led large-scale data curation, reproducible benchmarking, and baseline ASR development across nine Indian languages.
Developed a multimodal fusion model for joint ASR and dialect identification (DID), reaching 4.65% CER and 81.63% DID accuracy, integrating speech and text embeddings for Indian dialects.
Led the first large-scale study on DID in Indian languages (79.81% accuracy across 33 dialects).
Enhanced low-resource ASR via adapter-based domain adaptation, reducing WER by 10–30% and CER by 3–18% across nine languages.
Proposed a Gated Multi-Encoder ASR model with frame-level gating, reducing WER by 7.5% (Bhojpuri) and 9% (Bengali).

SYSPIN: Text-to-Speech Synthesizer in Nine Indian Languages (funded by GIZ, Germany)

Developed a data filtering pipeline using Kaldi-based ASR models, improving corpus quality.

Machine Learning / Signal Processing Consulting Engineer
ARTPARK @ IISc, Bengaluru, India
September 2022 – July 2026

VAANI: Capturing the language landscape for an inclusive digital India (funded by Google)

Co-author of the VAANI paper, to appear at NeurIPS 2026 (Evaluations and Datasets Track).
Developed automated signal processing and machine learning pipelines for acoustic quality assessment, speech activity detection, and metadata validation.
Established quantitative acceptance criteria and automated corpus curation protocols underpinning the open-source release of 31,000+ hours of speech from 156,000+ speakers across 109 languages and 31 states and union territories.

Undergraduate Student Researcher
Department of ECE, NIT Patna, India
July 2019 – June 2021
Advisor: Dr. S. Shahnawazuddin

Investigated formant-modification and bispectrum-based front-end features to improve children’s ASR under zero-resource conditions.
Developed a CGAN-based voice conversion framework, transforming adult speech into child-like speech for ASR adaptation.

Summer Intern
EICT Academy, IIT Guwahati, India
May 2019 – June 2019

Designed Verilog code for a 5-stage discrete wavelet transform cascade filter bank for FPGA-based simulation.

Reviewer

Conferences and Workshops: ICASSP (since 2025), IJCNN (since 2025), IEEE ASRU 2025, Interspeech (since 2026), IEEE SLT 2026.
Journals: Computer Speech & Language (Elsevier), Speech Communication (Elsevier).

Organizer

ASRU 2025 (MADASR 2.0): Lead organizer of the Multi-Lingual Multi-Dialect ASR Challenge in 8 Indian Languages; designed challenge tracks, evaluation protocols and public leaderboards, developed ESPnet-based baseline systems, and reviewed challenge track submissions.
ICASSP 2024 (LIMMITS’24): Co-organized session on Indic TTS, developed baselines, and reviewed challenge track submissions.
ASRU 2023 (MADASR): Led data curation and ASR baselines for Bengali and Bhojpuri.
Interspeech 2022 (Gram Vaani ASR): Organized the Hindi ASR challenge and developed Kaldi baselines.

Student Volunteer: ICASSP 2024, ICASSP 2025.

SLT-CODE Hackathon Winner 2022 — Won the Best Hackathon Project Award for developing dialectal ASR systems for Bengali and Bhojpuri, presented at SLT 2022.

A full list is available on the publications page and on Google Scholar.