Thanks for visiting my page.
People usually address me as Praveen. I am currently a Ph.D. student advised by
Prof. Mitesh Khapra
in the Department of Computer Science and Engineering,
AI4Bharat,
IIT Madras.
I completed my B.Tech in Computer Science and Engineering in
IIT Dharwad,
in 2021.
I'm currently working on building Neural Speech Synthesis systems with a focus
on enabling them to sound more natural, more intelligible, and have rich prosody while supporting multiple voices and styles,
for the major languages spoken across India.
In 2023, as a first step, we open-sourced SOTA TTS for 13 Indian languages
(demo).
In 2024, we focused on (i) making our models more intelligible
(IndicOOV,
ELAICHI)
, (ii) releasing the first-ever high-quality multilingual expressive dataset
(Rasa)
for any Indian language, and (iii) unlocking a massive speech corpus covering over 10,496 speakers
(IndicVoices-R)
that enables scaling Indian TTS!
In 2025, we scale our HQ studio recording efforts to 750 hours, covering 28 speakers across 16 Indian languages
(Rasa🤗).
We release the first instruction-following dataset
(RASMALAI)
for Text-Instruct TTS in Indian languages and a multilingual extension of Parler-TTS developed in collaboration with HuggingFace
(Indic Parler-TTS).
As we improve our data and models, we note that TTS evaluation remains tricky.
In our MUSHRA studies, we find biases—like reference matching and judgment ambiguity—that can distort progress tracking, and
we propose refined variants to mitigate them
(Rethinking MUSHRA).
While subjective assessments are good indicators of relative progress, they don’t test TTS systems in their real-world use - in isolation, where the goal is to sound human. We introduce the Human Fooling Rate
(HFR)
metric and a fine-grained variant for deployment-centric tracking with actionable insights.
(16/11/2024) Super grateful to receive the Google PhD Fellowship in 2024. Thank you
G
o
o
g
l
e
!
(15/11/2024) Recognized as an
Outstanding Reviewer in EMNLP 2024
.
My undergraduate journey has been detailed
here.