Namaskar Bengaluru, welcome to the ElevenLabs Summit. It's great to be back. India has been at the frontier of voice experiences, a truly voice-first country, and a part of ElevenLabs' story from the very start.
Some of our very first customers were here, and when we launched our very first model, we made sure it spoke Hindi. Today, India is our biggest market outside of the US, and it's always energizing to see what Indian businesses are building with our technology, a glimpse into the future that inspires many of the products we build.
Since ElevenLabs was founded, most of our industry has focused on solving intelligence. But as AI gets more capable, the bottleneck for adoption will become how well it interacts with people. Your AI system can be extremely powerful, but if it sounds robotic, it misunderstands or behaves strangely, people just will not trust it. And we think of it as IQ and EQ. AI is increasingly intelligent, but in much of the economy, interpersonal skills matter just as much as raw reasoning power. And you need both to create incredible experiences. People should be able to interact with AI as naturally as they interact with each other, through conversation and through voice, in their language and dialect, and of course, according to the social norms of a given place.
India is one of the most exciting markets in the world for ElevenLabs because the interaction is so complex. This is a country with 22 official languages, 122 major ones, and thousands of dialects and accents. How you speak is shaped by the region and the relationship. So if we can solve how people interact with AI here, well then I think, and I'm confident, we can solve it anywhere in the world.
Our goal is to build AI that is natural, intuitive, and trusted in everyday interactions, so everyone can benefit from this technology. But sounding human is just the first step. AI also needs to listen, understand the context, respond naturally, follow guardrails, and act on intent. Let me show you how we approach this.
ElevenLabs is first and foremost an AI research lab. At the core of the company is our audio AI engine, a suite of the industry's leading models across text-to-speech, speech-to-text, orchestration, dubbing, and even music. The models can speak, listen, and translate at human level.
Today I want to show you a few of these new models, starting with our latest model, Dubbing V2. Well, some of you, many of you, might know that the inspiration for ElevenLabs came from where I'm originally from, from Poland, where a foreign movie dubbed into Polish is dubbed by a single narrator. So the same voice for every character, whether it's male or female, and every emotion. So all the original incredible experience effectively disappears. We are a step closer to solving this. I recently had a chance of joining a podcast where we spoke with Nikhil about some of the work. Well, you will hear this in a second in English, that's the original. Then I will flicker, and you'll see how this was dubbed with our technology, with human-in-the-loop, to make sure we could bring that into a few other languages as well. Okay, without further ado, let me...
>> Do you think the medium of interacting with tech will move from hardware, which does feel unnatural to hold a thing...
Robots. Very quick. Tech.
It's always a little bit surreal to hear myself speaking all those languages, but judging by the reaction, it sounds like I can actually do it pretty well. But that's a step in actually really breaking down the language barrier, and something that we are so excited is actually possible. When we started the company, one of our very first customers was GeoStar here in India, where they were dubbing some of the incredible cricketers across IPL into all the different dialects across India, and that part of the story was a big part of ElevenLabs. And happily, all the way to today, we can bring the Dubbing V2 model. And the reason Dubbing V2 is special is that it can hear the original performance. So it carries the emotion, it carries the energy, it carries the emphasis as you move to the new language, and it can do it at this incredible level of quality. Next...
I want to show you Eleven v4, which we just released last week. It's our fastest and most emotive text-to-speech model on independent benchmarks and in quality. But to actually hear it, I will show it to you live. At ElevenLabs, we really like to show, not only tell. So, we'll swap to the laptop. So, here I prepared a simple sample. We'll go ahead and play it. As you can see, there's a good amount of directions of how the speech should be performed, with standard English and one of the voices.
Good morning, Bangalore. Okay, now things get a little more interesting. As you can see, I'm not just reading the words. I can change how I say them. I can slow things down or suddenly bring the energy right back up. The useful part is that all of this comes from the text itself. So I can generate speech, expression, reactions, and even little moments of sound without pre-recording any of it.
So you get the quality. But not only the quality, this final piece that was missing across AI models, the control, is finally possible. You can cue it exactly the way you want it, so the performance of the speech follows that direction. For any enterprise case, any deployment case, whether that's in a spoken word or in a conversation, this is essential.
I will now show you another sample where, to continue the theme, we'll try... And maybe before I do that, just a quick note: the voice you heard, which is incredible, is today with us in the room as well. It's one of the voice actors that is partnered with ElevenLabs, created their AI voice, shared it in the voice library. Every time it's being used, there's a compensation as a result, including the time we play. And it's one of our most incredible voices used across India. So if I can ask for applause for Raju, who is here with us as well.
Thank you, Raju. So now what we'll try to do is use my voice and see whether I can live speak a set of different languages.
Okay, now let's take it up a notch. Bangalore.
>> Kubaloo. You could see how it swapped automatically between all the different languages. Today, Eleven v4 supports 14 of India's languages, and we are planning to extend it to all 22 in the months to come. If we can swap back to the screen.
The reason it can do all of that is that v4 is built on an entirely new architecture, which delivers incredibly emotional speech in over 90 languages. Those of you who are building voice agents have probably always had to choose between whether it's fast or expressive. And now with Eleven v4 Turbo, you can have both, optimized for conversational use cases with 100 millisecond median latency.
Well, and we solved transcription too, even for the hardest of cases. Some of you might remember the thing behind me, which is the iconic Cello Butterflow ad. And what we'll do, you have the ad on the left. On the right, it will real-time transcribe what is being said in that video. And let's watch whether it was able to manage doing that.
Well, it got the name right, one of the longest names in existence, and happily was able to manage even with that task. But maybe on a little bit more serious note, we can fine-tune models to any specialized domain too. And recently we deployed a market-leading healthcare transcription model, with many more domains to come.
And while models are the foundation to apply AI and deliver its value out in the world, you need the full stack. So on top of leading models, we built Eleven Creative for storytelling and content production, and Eleven Agents, a full interaction platform for AI.
Today I want to focus on Eleven Agents, where we see the biggest shift in how we work and how work is done across businesses and organizations. This is where your product, operations, and customer support teams can create, run, and monitor human-sounding conversational agents across over 90 languages with that enterprise-grade security and control. It combines the seamless orchestration, configuration, integrations, and monitoring in one place. And it's perfectly optimized to work with our models, which improves the expressiveness, the latency, the reliability of agent interactions.
Our agents now handle more than 15 million conversations every week, and that's up three times in the last six months. But each deployment needs to be custom-built for any given industry, and that's exactly what we are doing and what we've done over the last years: specific interaction patterns, specific nomenclature, guardrails, compliance standards, with verticalized agent solutions, whether that's in healthcare, in government, in retail, and telco.
For established companies, there's a lot of messy and unglamorous work required to get AI into production on mission-critical tasks, and that's where our forward-deployed engineering teams come in. They are based here, based in the region, work side by side with you, are deeply embedded in the business, combine the domain expertise with the technology to help deploy faster. And because we work and own the full stack, we can adapt technology to your specific needs rather than just polishing something designed for an average company in a different industry.
Well, but we count ourselves extremely lucky to partner with so many of India's leading enterprises, creative and cultural organizations. One of the very first is Pocket FM. It's one of our first customers here in India and is creating some of the most incredible audio experiences. If you haven't, you need to download their app.
Razorpay uses Eleven Agents to allow millions of merchants on the platform to deploy conversational AI across subscription renewals, card recovery, customer engagement, and more across the full cycle, with that context preserved.
That point I particularly love in the work of what it means for the future with PhysicsWallah, because it's so important for how education and learning will evolve. So imagine if a student gets stuck late at night, they don't have to wait until the morning to get help. They can have a teacher they can talk through with PhysicsWallah, at their own pace, in their own language, with their own understanding of the subject, so they can improve and move a lot quicker.
And at Hubble, you can automate your home to respond to you, whether it's turning down the AC or switching on the geyser. Voice mode now supports eight languages in the Hubble app, which now has over 2.7 million downloads.
But alongside our work with businesses, we continue investing deeper across India, supporting developers with grants, partnering with governments, and working with nonprofits to make AI accessible to everyone. And to give you just two examples, we are working closely with the Indian government's Department of Empowerment of Persons with Disabilities to help restore the voices of people who have lost them due to illness or injury. And later today, we are formalizing a partnership with the government of Karnataka to bring safe and responsible AI to citizen services.
And this is just the start.
>> We'll continue, and we are investing in local teams across India so we can work more closely with organizations like hopefully all of you, that really are at the forefront of how AI interacts with people and how we are all changing how it interacts in the future.
