Anurag Kumar

Research Scientist, @GoogleDeepMind | Prev: @AIatMeta | CMU @SCSatCMU | @IITKanpur | Audio/Speech, Multimodal AI

[Anurag Kumar @AcouIntel]

Looking forward to @NeurIPSConf #NeurIPS2024 next week, I am there from Dec 11th-15th. Join our Audio Imagination Workshop on Dec 14th for engaging discussions on all things in audio generation space. We have an exciting list of papers and speakers. https://t.co/AwFxXg4kZ8

[Anurag Kumar @AcouIntel]

New work led by @realsonalkumar on bringing audio spatial intelligence to llms!

[Sonal Kumar @realsonalkumar]

Audio LLMs are good at what they hear, but mostly not where it's coming from.

From my @Google internship: SEA-LM, egocentric spatial audio understanding for smart glasses.

One model → what, where, when, who (you vs. bystanders), and what was said.

https://t.co/polowfIMw4

[Anurag Kumar @AcouIntel]

Phasecoder: Microphone Geometry-Agnostic Spatial Audio Understanding for Multimodal LLMs

Our @icmlconf #ICML2026 paper PhaseCoder, led by @artem_dementyev , takes a step toward enabling spatial hearing 👂 in LLMs and make them hear the way humans do. https://t.co/VwLXw4zyH7

While current LLMs can see images, read text, and transcribe speech, they miss a basic cue humans rely on constantly: where a sound came from. PhaseCoder enables AI to not just hear words, but to also understand where the world is.

The key thing that’s been missing is a representation that lets an LLM reason over direction, distance, speaker position, and “who said what from where.” Spatial hearing requires audio capture through multiple microphones, the geometry of which can vary. Phasecoder presents a spatial audio encoder that is agnostic to microphone geometry and converts raw multichannel audio into "spatial audio tokens".
It is not just doing source localization. We show that Gemma 3n fine-tuned with Phasecoder's "spatial audio tokens" can do complex spatial reasoning and targeted transcription tasks. It can answer questions like:

- Where is the speaker?
- Is the speaker on the left?
- What is the person farther away saying?
- Transcribe the person at 30°.

@artem_dementyev is at #ICML2026 @icmlconf to talk about this exciting new piece of work. Paper with @artem_dementyev, @wazeerzulfikar, Sinan Hersek, @PGetreuer, @vivek_kumar

[Sundar Pichai @sundarpichai]

Gemini Omni doesn't just build scenes that look real, it reasons about what should happen next. It combines an intuitive understanding of physics with Gemini's knowledge of history, science, and cultural context.

Rolling out today starting with video outputs to Google AI Plus, Pro and Ultra subscribers globally through the @Geminiapp + Google Flow, and @YouTube Shorts this week.

[koray kavukcuoglu @koraykv]

Today at Google I/O, we introduced Gemini 3.5 Flash! It has become an integral part of our daily research cycle and works with all the tools we have at Google.
We used a team of agents in Antigravity 2.0 to recreate the original AlphaZero research paper and build a playable version. They coded the reinforcement learning pipeline in JAX/Flax, trained a ResNet model from scratch via self-play on multi-TPU pods, and shipped a full-stack web app so you can play against it, from just 2 prompts. .
Here’s what else makes 3.5 Flash special 🧵

[Jeff Dean @JeffDean]

1/ Today at #GoogleIO, we’re releasing Gemini 3.5, our latest family of models combining frontier intelligence with action.

We’re starting by releasing 3.5 Flash, which is built to help you execute complex, long-horizon agentic workflows.

Gemini 3.5 Flash is our strongest model for coding and agent https://t.co/m62cBJhIjJ outscores 3.1 Pro on agentic and coding benchmarks like Terminal-Bench and MCP Atlas, while running 4x faster than other frontier models.

Used in Google Antigravity, 3.5 Flash is even further optimized to be up to 12x faster. It’s a powerful engine to deploy sub-agents that collaborate, run high-frequency iterative loops, and solve real-world problems at scale.

Some highlights we’re excited about 🔽

[Anurag Kumar @AcouIntel]

We are looking for reviewers for @ieeeICASSP 2026 for AASP areas. We received quite a bit more papers this cycle. If you don't currently review for ICASSP please consider doing so. Fill out the form below https://t.co/Vbv98lHogX

[Chenda @chenda54]

🚀 Join the ICASSP 2026 URGENT Challenge! Advance Universal, Robust & Generalizable Speech Enhancement.
🗣 Track 1: Universal Speech Enhancement
🎧 Track 2: Speech Quality Assessment

🔗 https://t.co/bZ3edVhIGM

#ICASSP2026 #SpeechEnhancement #AI #AudioProcessing

[Google DeepMind @GoogleDeepMind]

An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇

It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵

[Anurag Kumar @AcouIntel]

Check out the paper on Fri, Jun 13, ExHall D, evening session @CVPR #CVPR2025. Paper https://t.co/8L0NLvv6jY

[Anurag Kumar @AcouIntel]

Couple of papers at @CVPR #CVPR2025 (1) VisAH: Learning to Highlight Audio by Watching Movies. How do you transform a poorly mixed audio into a well-balanced audio ? VisAH learns to leverage visual cues by training from movies which naturally provides the required supervision

[Anurag Kumar @AcouIntel]

(2) XRIR: Hearing Anywhere in Any Environment. A key problem in neural RiR estimation has been cross-room generalization. We make an attempt to address this and introduce a large scale dataset ACOUSTICROOMS, with 300,000 high-fidelity RIRs simulated from 260 diverse rooms.

[Nando de Freitas @NandoDF]

RL is not all you need, nor attention nor Bayesianism nor free energy minimisation, nor an age of first person experience. Such statements are propaganda.

You need thousands of people working hard on data pipelines, scaling infrastructure, HPC, apps with feedback to drive benchmarks and data, tons of research and engineering on generative models, data mixtures, ablations, RL/selftraining, etc etc and we will probably need lots of people working hard to figure out safety, causal world models, awareness, models that create abstractions comparable to infinity and zero and use these to predict the existence of things like black holes and suggest experiments to verify such hypothesis, or come up with novel engineering designs to generate energy more efficiently, robotics, etc etc.

It takes thousands of people and many ideas. In the end some simple ideas might become obvious but such obviousness only happens in retrospect. Yes, there is a bitter lesson but if we had followed it, we’d still be doing linear regression with RL. Let’s not oversimplify, but rather honour the research and engineering of thousands of people.

Also, people keep rewriting history. When our language understanding start up (darkbluelabs) was acquired by Google about 10 years ago, we joined DeepMind, where the AGI documents were all about concepts, RL, episodic memories and made it clear that there was no room for language. To be honest, back then such a position wasn’t so crazy. Now it seems silly, but only because of the benefit of hindsight.

There’s no 1 or 10 heroes in the history of AI. There’s many 1000s of hard working students, profs, engineers, operations and support people, product folks, managers, even hedge funds among others. Let’s honour the whole community and not just ceos or the philosophers of Bayes, RL, deep learning, etc. I look forward to learning from the next generation and seeing what they will achieve. To them: Don’t buy the existing narratives blindly, innovate. Remember that just like mathematics, AI will advance one grave at the time.

[Anurag Kumar @AcouIntel]

(2) Reexamining the Efficacy of MetricGAN for Speech Enhancement. Led by @realHaibinWu. Showcases some crucial limitations of MetricGAN, and proposes some training tricks to address. (already presented, but check out the paper) https://t.co/lNouyTwyka (3/3)

[Anurag Kumar @AcouIntel]

@ieeeICASSP is finally happening at a place for which I don’t need a visa to travel 😀, but not able to attend this year #ICASSP2025. If you are there, check out these two papers I co-authored. (1/3)

[Anurag Kumar @AcouIntel]

(1) Advancing Active Speaker Detection for Egocentric Videos. Led by @huh_jaesung. SOTA for active speaker detection in challenging ego-centric videos. Session: Machine learning for multimodal data I Apr 11: 11:30 am - 1:00 pm. https://t.co/CCvoJ53nT6 (2/3)

[Anurag Kumar @AcouIntel]

Career Update: Excited to join Google Deepmind @GoogleDeepMind to continue working on audio/speech/multimodal AI.
I left Meta @Meta after more than 6 years and I will definitely miss working with some amazing friends and colleagues. Super thankful for all the fun collaborations.

[Shrestha Mohanty @shremoha]

So happy to share that our work has been accepted to
@SIGIRConf. Thank you to my amazing collaborators! @NegarEmpr, Andrea Tupini, Yuxuan Sun, @Tviskaron, @artemZholus, @Cote_Marc and @julia_kiseleva
Pre-print: https://t.co/ctdcHZTVqw

[arXiv Sound @ArxivSound]

``Efficient Audiovisual Speech Processing via MUTUD: Multimodal Training and Unimodal Deployment,'' Joanna Hong, Sanjeel Parekh, Honglie Chen, Jacob Donley, Ke Tan, Buye Xu, Anurag Kumar, https://t.co/29Yg1je6zB

[Anurag Kumar @AcouIntel]

The paper explores how LLMs can be used to effectively contextualize excerpts from conversations to improve understandability, readability, and other factors and reduce misinterpretations.

[Anurag Kumar @AcouIntel]

Exciting new work focusing on comprehension of long-form social conversations @coling2025 #COLING2025. https://t.co/hpO8TFn4Bg. All thanks to the hard work of @shremoha.