Camb AI

Introducing MARS8 - the first family of Text-to-Speech AI models by the research team at CAMB AI Because voice AI needs to serve 8 billion people, and one model cannot serve every use case.

PSG vs OM. Le Classique. Earlier this year, we powered the Trophée des Champions with live AI-dubbed commentary, a first for European football. It was so nice…so we had to do it twice. Last month, during the @Ligue1 encounter between the two sides, we delivered AI-dubbed Italian commentary live, translating the French broadcast in real time. Just a new language layer without any additional production workflows. For leagues, that means reaching new markets that were previously too complex or costly to serve. Keen to see it in action? Watch the clip below @LFPfr

The AI world gets together in San Jose this week 🇺🇸 @NVIDIAGTC is back for another edition, bringing together researchers, builders, and companies shaping the next generation of AI. Felix Fredriksson from our team will be there on-site to connect with like-minded builders and showcase our localization infrastructure For us, GTC is a reminder of what we’re building: the voice and language infrastructure that allows content to travel across languages while preserving the original voice, emotion, and nuance. If you're attending GTC this week, reach out to Felix to learn more about our enterprise solution

Not all voice belongs in the cloud MARS-Nano is the smallest model in our MARS8 family of TTS models, designed for high-quality on-device voice. It runs in constrained environments while preserving naturalness and clarity. Expressive voice that runs entirely on-device, from phones to embedded systems, bringing lower latency, stronger privacy, and reliable performance even when connectivity isn’t guaranteed. Voice AI just became a core part of on-device infrastructure.

Read the full announcement here - camb.ai/blog-post/euro…

The Paralympic Winter Games deserve to be experienced by everyone! We’re proud to share that the European Broadcasting Union @EBU_HQ and CAMB.AI will deliver live and on-demand subtitling for the Milano Cortina 2026 Paralympic Winter Games, taking place from 6–15 March 2026. Across all six Paralympic Winter sports streamed on Eurovision Sport, audiences will be able to follow the action with real-time subtitles, powered by our AI-driven speech-to-text infrastructure. For a global event that represents global unity, accessibility matters. Subtitles help ensure more people can follow the action live, in sound-off environments, or on-demand We’re happy to join forces with Eurovision Sport again, following our collaboration that began in 2024, marking another step forward in making major international sport more accessible at scale. Learn the full scope of what we will deliver together with Eurovision Sport in the announcement below

What a day 1 at #Sportsinnovation 🚀 With the event in full swing, Marte Harkestad took part in a panel discussion alongside Johannes Franken, Luke McCoy and Jean-Baptiste Alliot, illustrating what makes our localisation infrastructure a valuable asset for sports rightsholders in optimising operations and reaching newer audiences across the world A whole day 2 ahead of us. If you’re still around, drop by and say hi to Marte and the team at booth P3!

Check out the full report by SportsPro - sportspro.com/projects/globa…

It’s always a good feeling when your work gets recognised by the industry you’re building for Proud to announce that CAMB.AI has been featured in @SportsPro's Global SportsTech Report 2026 as one of the companies to watch Sport was the hardest problem to solve for us Live commentary, crowd noise, sudden shifts in emotion, overlapping voices, capturing all of that, then transferring it across languages without losing tone, identity or intensity, is one of the toughest challenges in AI localization And that’s the challenge we chose to take on from day one. From live AI-dubbed commentary to multilingual VOD and archive content, our goal has always been to enable sports organisations speak to global fans in their native language, without compromising quality or speed. Something we have enabled for the likes of Major League Soccer, European Broadcasting Union (EBU) Sport, NASCAR and other major rightsholders across the globe A big thank you to the SportsPro team for the mention If you’re curious about where media, content and AI are heading in sport, the full report is worth a read (link in comments)

What a special way to end February! Last week, Our CTO and Co-founder, Akshat Prakash, took the stage at the Aegis Graham Bell Awards (AGBA) for Innovation Exhibition in New Delhi, an event backed by the Government of India: Ministry of Electronics and Information Technology, celebrating breakthrough innovation across AI, deep tech and enterprise transformation. Akshat spoke about something we deeply care about at the ‘Innovation Talk’: Building localization infrastructures as a core foundation that makes enterprises truly accessible across languages, markets and audiences What made AGBA even more special was being recognized as a winner in the ‘Innovation in AI - Enterprise’ category For a company that started just a few years ago, being evaluated alongside some of the biggest, most established technology players in the ecosystem is a big testament to what we’re building. A nice reminder that impact isn’t defined by size but by the right ambition and the problems you choose to solve. A big thank you to the Government of India and the Ministry of Electronics & Information Technology, for hosting such a strong gathering of builders, researchers and operators shaping India’s innovation ecosystem. We look forward to being back!

We’re heading to SportsInnovation 🇩🇪 On March 4–5, CAMB.AI will be in Düsseldorf for SportsInnovation 2026, one of Europe’s leading invite-only events where sport, media, and tech decision-makers come together to explore the future of the sport industry. Organised by the DFL Deutsche Fußball Liga, SportsInnovation takes place every 2 years and is shaped around live showcases, real match scenarios, real production environments, and real innovation. Our CEO Avneesh Prakash and Marte Harkestad will be in attendance. If you happen to be in attendance, drop by booth P3 to know more about how we have catered to major leagues around the world like the LFP - Ligue de Football Professionnel and Major League Soccer. And don’t miss Avneesh’s panel on Day 1 on the impact of off-the-pitch technologies in scaling broadcast workflows and efficiency for rightsholders See you in Düsseldorf!

Sometimes “good enough” isn’t enough. MARS-Instruct was built for movie studios and production houses that need precision. While Flash is built for real-time conversations, and Pro is meant for live media broadcasts, Instruct lets you control how a voice sounds and feels in premium dubbing workflows for movies and television Prompt MARS-Instruct to fine-tune emotion, tonality, breathing and every subtle detail of the performance, until it fits the scene exactly as you imagined it. Director-level fine-grained control for film, high-end TV, and premium dubbing workflows. Go explore MARS-Instruct today

Run the benchmark yourself! - github.com/Camb-ai/MAMBA-… View benchmark results of MARS8 - #benchmarks" target="_blank" rel="nofollow noopener">camb.ai/mars#benchmarks

Voice AI sounds impressive in a demo But in a real-world production scenario, you get 2 seconds of audio Different language. Background noise. And you still need the voice to sound right So we benchmarked MARS8 in those conditions, not in a studio. Our open MAMBA benchmark uses: 1,334 samples 70% cross-language cloning 2.3s average reference length The result is that our family of TTS models, MARS8, leads in speech quality. 5.67% CER on multilingual content. 0.87 speaker similarity (WavLM). 0.71 on CAM++ — (a 38% jump over the next best model) All from ultra-short, expressive audio If you’re building with voice, this is what production-grade actually looks like. And yes, the benchmark is open. Go check it out today 👇

Go deploy on Google Vertex AI today - console.cloud.google.com/vertex-ai/publ…

MARS8 is now live on @Google Vertex AI! Why this matters: Voice is increasingly becoming core to AI products, not as an ad-on feature that can be integrated later but as part of the overall infrastructure. By being live on Vertex AI, MARS8 offers ultra-low latency, the ability to privately deploy within your own environment, meet data residency and locality requirements, and operate on flat, predictable costs, unlike other usage models that would be restrictive to scale. For teams building agents, media tools, or global products, production-grade multilingual speech is now available where your AI already lives. Go deploy MARS8 today on Google Vertex AI 👇

Check out the LiveKit and LangChain documentation: docs.livekit.io/agents/models/… docs.langchain.com/oss/python/int…

CAMB.AI Now Integrates with @LangChain & @livekit If you’re building agentic or real-time voice applications, this makes things a lot simpler CAMB.AI can now be used natively inside LangChain workflows and as a drop-in TTS provider for LiveKit agents, allowing developers to deploy MARS8 models in any workflow without reworking their existing architecture. With Langchain and Livekit, our voice models fit directly into modern agent stacks and live systems, whether you’re building tools or deploying real-time, interactive agents. Explore the Langchain and Livekit integrations below👇

Try CAMB.AI for Live - camb.ai/streams And check us out on Product Hunt - producthunt.com/products/dubst…

Live dubbing was once reserved for the world's largest broadcasts. That era is over Today, we’re making DubStream available to everyone Over the last 2 years, our live dubbing tech has powered some of the most demanding live environments, from @NASCAR races streamed in Spanish to the @MLS All-Star Game broadcast in 4 languages and live demo days and events translated in real time. Starting today, that same technology is open to any creator aspiring to reach global fanbases, sports leagues competing globally, and (not limited to) educators delivering live lessons in the language their students understand. 80% of the world doesn’t speak English as its first language. Your live content should reach all of them. DubStream is also live on @ProductHunt today 🚀 We’d appreciate your support. Link in comments 👇

We built Flash for ultra-low latency voice that holds up in real conversations, where response time is part of the experience. Running on Baseten means deploying close to users, avoiding hard concurrency caps, and scaling without unnecessary network hops. If you’re building conversational AI and latency matters, this is worth a look. Get 5 hours of free MARS-Flash usage on Baseten today - baseten.co/talk-to-us/mar…

How crucial is response time for voice agents? It’s the difference between flow and friction in a conversation. When responses lag or agents get interrupted, the experience breaks and being limited by tokens doesn't help either. That’s exactly the problem we set out to solve. We’re delighted to announce that MARS-Flash is now LIVE on @baseten