Interspeech 2026

My first big academic conference, in person

For someone who has been in academia and research, attending a lot of academic conferences would be normal. That was not the case with me.

The conferences I did attend were either not the big-name events, or did not come directly from my career. The ones on scripts and fonts, like Typoday 2018 and Grafematik 2018, came from projects outside my work and research.

During my PhD, when conferences should have been a regular part of life, things were not in my favour. It was the COVID years, and every conference happened online. Even the big one, EMNLP 2024, I attended online, because of the short notice and visa and funding issues.

So attending Interspeech 2026 in Sydney in person was a really big thing for me. People might not even believe it if I said this was my first time travelling abroad alone.

/img/interspeech-2026/sydney-selfie.jpg

Our team from Adalat AI presented two oral papers: Vividh-ASR, a complexity-tiered benchmark for robust Indic speech recognition, and SCRIBE, on diagnostic evaluation and rich transcription models for Indic ASR. SCRIBE looks beyond a single word error rate, breaking mistakes down into everyday words, numbers, punctuation and field-specific terms, so you can see where a model actually needs work.

Alongside the papers, we shared two releases: Koyal, our open speech recognition models for Hindi, Kannada, Malayalam and Telugu, and scribe-eval, the diagnostic evaluation framework behind SCRIBE.

Even if you are not someone who starts conversations, it was easy, and exciting, in an environment where everyone has something to share about your area of interest. The person chairing your session might have something specific to add to what you were talking about.

I was so privileged to have Tanja Schultz, former ISCA president, chair my session. She has worked on Malayalam-English code-switched speech. I also got to meet Roger K. Moore and thank him personally for his Speech 101 talk. Because of it, my thesis has a full section on grapheme-to-phoneme conversion in Malayalam that does not confuse phones with phonemes.

/img/interspeech-2026/collage.jpg

Not everyone I had a good conversation with made it into the pictures. I met many old friends, former collaborators, and people I have admired and cited in my work. It was lovely to talk with Prachi Singh, Sriram Ganapathy, Jesin James, Kathy Reid and Aravind Krishnan. I also met Shangeth and Abhinav, Manas’s former colleagues and wonderful people. Thank you, Manas, for the introduction. I had some really critical conversations with Sanjeev Khudanpur. I also found some really cool work on tokenizers and their impact on Indic language LLMs and speech, SᴜTRA, and had good conversations with its authors.

It really means something to be there in person, even though it is not always easy, or possible.

Outside the conference

Yes, it was my first time seeing the Pacific. I met the animal with a pouch, and penguins too. The Opera House and the Harbour Bridge got ticked off as well.

And I spent one good day with a person from my school who really had a role in shaping who I am as a person. Thank you, Athira, for the coastal walk.

/img/interspeech-2026/coastal-walk.jpg

Back home

I know it was not easy at home for Santhosh, juggling work and Vani’s school days. This was my longest time away from Vani in six years. Amma came over to support them, and all of it let me roam guilt-free during these days. Thank you!!

/img/interspeech-2026/postcard-from-sydney.jpg

See also

comments powered by Disqus