Hi, I’m Dia Krishnan (aka Aditya Srinivas Menon), a Machine Learning Researcher at Adalat AI. I build speech recognition and speaker diarization systems for low-resource Indic languages, designed to run in noisy courtrooms. My work spans multilingual and code-switched ASR, streaming diarization and efficient fine-tuning of speech models. You can learn more about my research themes here.

News

• Oct 2026 Preprint: Slots Are Not Speakers, a training-free speaker linker for streaming diarizers.
• Sep 2026 Released Koyal, open multilingual Indic ASR models.
• Jun 2026 Vividh-ASR accepted at Interspeech 2026.
• May 2026 1st place on both the diarization and ASR tracks of the DISPLACE 2026 challenge.
• May 2026 Presented Windowed SummaryMixing at ICASSP 2026 in Barcelona.
• Jan 2026 Listen Like a Teacher published at AAAI 2026 (Main Technical Track).
• Dec 2025 Joined Adalat AI as a Machine Learning Researcher.
• Aug 2025 LASPA published at Interspeech 2025.

Industry and research experience

At Adalat AI, I drove the migration from Whisper to faster in-house legal ASR models across English, Malayalam, Kannada, Odia and Hindi, now used in 4000+ courts across 11 states. The move cut inference cost by 60%, with 2.4x lower latency and 5x more concurrent users per GPU. I also built a diarization + ASR service for arbitration, designed the Indic ASR recipe behind our open Koyal models using an NVIDIA compute grant, and developed a causal speaker linker that lifts the 4-speaker cap of streaming diarizers.

Before that, I spent a year and a half at Sony Research India working on language-agnostic speaker representations, Whisper hallucination mitigation, efficient fine-tuning of SSL models, synthetic data for profanity detection in Hindi, Marathi and Bengali, and ASR, VAD and diarization systems for Indian languages.

My internships at Begin.ai, Bosch India and AdPushUp gave me a taste of the real-world applications of AI. I had the wonderful opportunity to be guided by Dr. Konjengbam Anand, who taught me a lot about research, cross-lingual NLP and low-resource languages, which led to the internship at IIT Dharwad and publications at SPECOM'23 and ICDMAI'23. I worked with Prof. Dr. Chris Biemann from University of Hamburg, Germany, on interpretable sentiment analysis models for low-resource languages, and spent the summer of 2023 on an end-to-end person re-identification pipeline with Dr. Deepak Mishra at the Indian Institute of Space Science and Technology (IIST). For a detailed overview, head over to my CV or download it!

Open source

• Koyal: open ASR models for Hindi, Kannada, Malayalam and Telugu, plus a 600M multilingual Indic model.

Education

I graduated with B.Tech with Hons. from IIIT Kottayam in 2024, with a CGPA of 9.01/10. My honours thesis was on a novel approach to cross-lingual phrase retrieval using deep learning, under the guidance of Dr. Manu Madhavan. My final year project was a study on building a web-based verb thesaurus for Sanskrit, a low-resource language in India, under the guidance of Dr. Nandini J. Warrier.

Talks

• ICASSP 2026, Barcelona: Windowed SummaryMixing
• The Fourth Indian Symposium on Machine Learning (IndoML): Cross-lingual Phrase Retrieval
• National Youth Conference on Indian Knowledge Systems: Language Technologies for Low-Resource Indian Tribal Languages
• Cyber Security Roadshow, Indian Institute of Science (IISc): Attack Prediction Using VICReg
• Machine Learning Research School 2023: NLP for Low-Resource Languages
• Argentine Symposium of Artificial Intelligence: Image Inpainting Applied to Art
• Hosted the N-Large NLP and Large Language Model workshop at IIIT Kottayam

A lil bit of boasting

• 1st place, diarization and ASR tracks (Phase 2), DISPLACE 2026
• Winner, Manthan 2021, India’s first hackathon on national security (BPR&D and AICTE)
• Finalist, Smart India Hackathon 2022, for 2 problem statements
• #9 worldwide among 65,000+ hackers, Major League Hacking LHD: Share 2021
• Winner (out of 600+ teams) at RU Hacks, Hack Tech and MLH: Uncommon Hacks
• Reviewer, AAAI 2026; Program Committee, N-Large workshop

When I’m not diving into complex algorithms, I mentor at hackathons and non-profits, organise technical events, and support underrepresented groups in STEM.