Satvik Dixit

Voice Agent Evals | Prev @CarnegieMellon @IITDelhi

[Thinking Machines @thinkymachines]

Today, we are introducing Inkling.

Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.

https://t.co/Ghebq5mG30

Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵

[OpenAI @OpenAI]

Introducing GPT-Live, a new generation of voice models for natural human-AI interaction.

Rolling out in ChatGPT starting today.

You’ll want to turn the sound on for this one.

[Karan Goel @krandiash]

We released Sonic-3.5 and Ink-2, the #1 streaming models for text to speech and speech to text you can use in your voice agents today.

New architectures enable new frontiers for speed and quality.

We're now the only provider to have #1 models for both speaking and listening.

[Chris Donahue @chrisdonahuey]

This one feels special. Finally we have real time on device generation w/ instrument-like (200ms) latency. Run it in your DAW or on an airplane! Proud to have been a part of this

[Satvik Dixit @SatvikDixit9]

Great turnout at our hackathon at the @ycombinator office today. Exciting to see all the interest in voice agents.

Thanks to everyone who joined us, and to @trydaily, @awscloud, @NVIDIAAI, and @twilio for helping make it happen.

[Sesame @sesame]

A collection of personal agents, crafted for everyday conversation.

Preview available now on iOS.

https://t.co/Hq5ZXAqPvX

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Satvik Dixit @SatvikDixit9]

Code for training mellow-like models released as well: https://t.co/v39lyz5ibg

[Satvik Dixit @SatvikDixit9]

Our paper "Mellow: a small audio language model for reasoning" was accepted to #NeurIPS2025!

Catch our poster in Session 3 at 11am today.

[Soham Deshmukh @sohamdesh_]

we show for the first time ever that sub-billion audio models can reason. we introduce mellow, a small audio-language model (167M) that gets SoTA on different audio reasoning tasks. by using our method and data, you can train an alm within 24 hrs on academic resources (1/n 🧵)

[arXiv Sound @ArxivSound]

Satvik Dixit, Koichi Saito, Zhi Zhong, Yuki Mitsufuji, Chris Donahue, "FoleyBench: A Benchmark For Video-to-Audio Models," https://t.co/cQDiSOeW2J

[Satvik Dixit @SatvikDixit9]

Excited to be at WASPAA 2025!

[Chris Donahue @chrisdonahuey]

Excited to share our beta release of Music Arena, a live evaluation platform for state-of-the-art AI music generation models!

🎧 Listen to the latest models and 🗳️ vote for your favorite

⚔️ https://t.co/XyXvzlOMcH
⭐️ https://t.co/9f3mvseEyu
📜 https://t.co/HqTg5PH09O

[Chris Donahue @chrisdonahuey]

Excited to announce 🎵Magenta RealTime, the first open weights music generation model capable of real-time audio generation with real-time control.

👋 **Try Magenta RT on Colab TPUs**: https://t.co/SofwwmWlxC

👀 Blog post: https://t.co/QTypsxNHvK

🧵 below

[Neil Zeghidour @neilzegh]

Thanks @GoogleAI 🙏, I'm proud to see concepts introduced in this paper (RVQ-VAE, quantizer dropout) being still as relevant four years later, and in particular how the RVQ turned out to be a perfect fit for audio language models.

[Satvik Dixit @SatvikDixit9]

Paper: https://t.co/7BLA1D8bdL

[Satvik Dixit @SatvikDixit9]

Excited to be at #ICASSP2025!

If you’re working on or interested in audio language models, feel free to reach out.

Also, come by our poster on audio caption evaluation at the SALMA Workshop tomorrow at 4 PM.

[Neil Zeghidour @neilzegh]

Trimodal training (text-audio-img) is challenging because you a have a lot of unimodal data, some bimodal and few to none with all 3 modalities & combining them is not obvious. We propose a simple extension to Moshi that allows it to understand images.

[𝚐𝔪𝟾𝚡𝚡𝟾 @gm8xx8]

Mellow: a small audio language model for reasoning

[arXiv Sound @ArxivSound]

``Mellow: a small audio language model for reasoning,'' Soham Deshmukh, Satvik Dixit, Rita Singh, Bhiksha Raj, https://t.co/2frf6F1gWq

[Neil Zeghidour @neilzegh]

Today we release Hibiki, real-time speech translation that runs on your phone. Adaptive flow without fancy policy, simple temperature sampling of a multistream audio-text LM. Very proud of @tom_labiausse 's work as an intern.