kavyamanohar.com • github.com/kavyamanohar • ORCID • Huggingface
Speech and Language Technology • Technical Education • Responsible AI
Machine Learning Researcher
March 2025 - Present : Adalat AI.
Computational Linguist
2023 - 2025 : Kerala University of Digital Sciences Innovation and Technology.
Assistant Professor in Electronics and Communication Engineering.
2017 - 2019 : Government Engineering College, Palakkad.
2013 -2017 : Aryanet Institute of Technology, Palakkad.
2012 - 2013 : Lourdes Matha College of Science and Technology, Thiruvananthapuram.
APJ Abdul Kalam Technological University, 2019-2023
Thesis: Linguistic challenges in Malayalam speech recognition: Analysis and solutions.
University of Calicut, 2010-2012
Communication Engineering and Signal Processing with CGPA of
8.16.
Government Engineering College Thrissur with MHRD Fellowship.
Thesis: A comparative study on vector quantization techniques on
the perceptual quality of wideband speech.
University of Kerala, 2006-2010
Electronics and Communication Engineering from Government Engineering College, Thiruvananthapuram with a CGPA of 7.6.
Model Evaluations:
Revealed critical gaps in OpenAI Whisper’s multilingual speech AI model evaluation routine through systematic analysis of text normalization for Indic languages, demonstrating performance overestimation due to unaccounted linguistic features. Research published at EMNLP 2024 Main Conference.
Subword Tokenization for Improved Speech Recognition Model Performance:
Developed a novel syllable-byte pair encoding (S-BPE) tokenization method that intelligently combines linguistic syllable boundaries with data-driven byte pair encoding for Malayalam speech recognition. Achieved 16.8% word error rate (WER) reduction over word-level baseline, demonstrating superior scalability with increased text only training data while maintaining computational efficiency during inference. Published in EURASIP Journal 2023.
Enhancement to Kaldi ASR toolkit:
Enhanced Kaldi ASR toolkit’s flexibility by implementing custom subword boundary support, contribution merged into main codebase.
Grapheme to Phoneme (G2P) System Development
Created and published Mlphon, an open-source Malayalam G2P conversion library with web interface, enabling automated pronunciation generation. It is now employed in the development of automatic speech recognition and speech synthesis systems. The work in published in IEEE Access Journal.
Extended Epitran multilingual G2P library’s language coverage by implementing Malayalam support.
Dataset Creation and Curation
First comprehensive Malayalam pronunciation dictionary, published on Hugging Face.
Curated and published crowd sourced Malayalam speech corpus, enabling broader ASR research for 35M+ speakers working alongside the FOSS Language Techology Collective, SMC.
Pioneered speech recognition development for endangered languages (Malasar, Poumai Naga), working closely with language community members through finetuning pretrained transformer models, enabling first-ever speech recognition capabilities. Publication at ICON 2023.
Boosted ASR accuracy by integrating language models with transformer acoustic models, demonstrating significant WER reductions. Models and Demo published in Huggingface. Detailed technical report and code published in personal blog.
Developed and deployed Malayalam ASR system with web-based demo with client side inference in connection with PhD research.
Linguistic Typology Studies
Published first quantitative analysis of Malayalam’s morphological complexity in comparison with Indian and European languages (TSD 2020), establishing empirical complexity metrics for cross-linguistic comparison.
Digital Typography and Font Engineering
Engineered OpenType glyph formation rules for three major Malayalam fonts (Manjari, Chilanka, Gayathri), enabling correct rendering of complex Malayalam scripts in digital platforms, now widely used across Kerala’s digital ecosystem.
Book
Book Chapter
Journal
Improving speech recognition systems for the morphologically complex Malayalam language using subword tokens for language modeling. Kavya Manohar, A. R. Jayan, and Rajeev Rajan. J AUDIO SPEECH MUSIC PROC. 2023, 47 (2023).
Mlphon: A Multifunctional Grapheme-Phoneme Conversion Tool Using Finite State Transducers. Kavya Manohar, A. R. Jayan, and Rajeev Rajan. IEEE Access 10 (2022).
Conferences
Romanized to Native Malayalam Script Transliteration Using an Encoder-Decoder Framework, Baiju, Bajiyo, Kavya Manohar, Leena G. Pillai, and Elizabeth Sherly. COLING IndoNLP Workshop 2025 (Accepted)
What is lost in Normalization? Exploring Pitfalls in Multilingual ASR Model Evaluations, Kavya Manohar, Leena G Pillai. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Miami, Florida. Association for Computational Linguistics.
Enhancing End-to-End Malayalam Automatic Speech Recognition with Language Model Augmentation. Kavya Manohar, Ashish Abraham, Gokul G Menon. Speech and Language Technologies for Low-Resource Languages. SPELLL 2024.
Malayalam to English Named Entity Transliteration using Attention based BiLSTM, Bajiyo Baiju, Kavya Manohar, Leena G Pillai and Elizaebth Sherly. IEEE-RAICS, Recent Advances in Intelligent Computational Systems at Kothamangalam, 16-18 May 2024
Automatic Speech Recognition System for Malasar Language using Multilingual Transfer Learning, Basil K Raju, Leena G Pillai, Kavya Manohar, Elizabeth Sherly. In Proceedings of the 20th International Conference on Natural Language Processing (ICON 2023)
Automatic Recognition of Continuous Malayalam Speech using Pretrained Multilingual Transformers, Kavya Manohar; Gokul G. Menon; Ashish Abraham; Rajeev Rajan; A. R. Jayan. 2023 International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
Syllable Subword tokens for Open Vocabulary Speech Recognition in Malayalam, Kavya Manohar, A. R. Jayan, and Rajeev Rajan. Third International Workshop on NLP Solutions for Under Resourced Languages (NSURL 2022), Trento, Italy.
Quantitative analysis of the morphological complexity of Malayalam language. Kavya Manohar, A. R. Jayan, and Rajeev Rajan. International Conference on Text, Speech, and Dialogue. Springer, Cham, Czech Republic (2020).
Malayalam Orthographic Reforms. Impact on Language and Popular Culture. Kavya Manohar and Santhosh Thottingal. Proceedings of Graphemics in the 21st Century, Brest, France (2018).
Spiral splines in typeface design: A case study of Manjari malayalam typeface: Santhosh Thottingal and Kavya Manohar. Typoday Conference, Mumbai (2018).
Comparative study on vector quantization codebook generation algorithms for wideband speech coding. Kavya Manohar and B. Premanand. 2012 International Conference on Green Technologies (ICGT). IEEE (2012).
Articles
Indian Languages and Text Normalization - Article Published in May 2024.
Natural Languages and Machine Intelligence - A Malayalam article published in Sasthragathi, June 2022.
Phonetic description of Malayalam consonants - A study of phonetic description of Malayalam consonants based on existing methods and IPA. Published on January 2020.
Language technology in the age of AI - A Malayalam article published in Janayugam, Sept 2019.
Information, Entropy and Malayalam - What is the information Entropy of Malayalam language? How to calculate it? Published on July 18, 2019.
u and u: vowel signs of Malayalam - An analysis of various visual forms of Malayalam u-signs, published at Alphabettes.org.
Diversity and Inclusion Grant from the Association of Computational Linguistics to present at EMNLP 2024.
Best Paper Award for the paper “An Open Framework for the Development of Automatic Speech Recognition in Malayalam”, at Kerala Science Congress 2023 organized by Kerala State Council for Science, Technology and Environment.
Junior Research Fellowship by University Grants Commission, Government of India. 2019-2023 (Ph.D.)
GATE Fellowship by Ministry of Human Resource Development, Governemnt of India, 2010-2012 (M.Tech.)
Member, Indian Language Technologies And Products Sectional Committee LITD 20, Bureau of Indian Standards.
Program Committee Member, SPELLL 2024, Third International Conference on Speech and Language Technologies.
Program Committee Member, ComputEL-8, Eighth Workshop on Computational Methods for Endangered Languages
Reviewer for SPELLL 2024, ComputEL-8, International Conference on Emerging Technologies for Intelligent Systems (ETIS 2025).
Summer School on Automatic Speech Recognition, IIT Guwahati, 2019.
Machine Learning by Stanford University on Coursera, 2017.
A System View of Communications: From Signals to Packets by Hong Kong University of Science and Technology on edX, 2014.
| 18 March 2025 | Kavya Manohar |
| sakhi.(myfirstname)@gmail.com |