Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

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.

Karan Goel

20,887 subscribers

7,128,131 views • 3 months ago •via X (Twitter)

Comedy Education Science & Technology

Anya Rossi• Live Now

Private livecam show

44 Comments

Karan Goel3 months ago

If you switch to Cartesia from your existing provider, three months are on us. Hear the difference today. 👇🏻

Karan Goel3 months ago

We’ve seen incredible traction this year, with enterprises transforming every part of their operations with Voice AI. We’re sharing what we’ve learned and are giving away a guide to the top 10 enterprise use cases for Voice AI. Retweet and comment "CARTESIA" on the main video and we'll send you the guide and $100 in credits so you can get started today.

Karan Goel3 months ago

Try a call with a real voice agent and sign up for the promo here!

Deedy3 months ago

its so hard to make a text-to-speech model with such a sub-300ms median ttft (time to first token). big big congrats!

kabir3 months ago

can't wait to fly on Air Cartesia 😍

Lenny Rachitsky3 months ago

Making the word more magical, one model at a time. Go Cartesia!

Karan Goel3 months ago

thanks Lenny, appreciate the support!

Nimit Sohoni3 months ago

bro is always stealing croissants

Guillermo Rauch3 months ago

Incredible

Timothy Luong (Chongz)3 months ago

Help help @krandiash's stealing croissants again

Dhwani Garg3 months ago

objectively the best team out there

Bucky Moore3 months ago

next level sweater game per usual

Karan Goel3 months ago

😅 running out of sweaters for these videos

Orikan3 months ago

Lol 11labs is so cooked, their speed is nothing compared to this

Brandon Yang3 months ago

congrats to the team!!

Baseten3 months ago

Congrats! 🚀

Karan Goel3 months ago

thanks!

shreya rajpal3 months ago

tts quality on the demo is smooth af, congrats on the release 🚀

Brandon Yang3 months ago

CARTESIA

Sathwik Tejaswi3 months ago

Not saying this coz it's our model but it's really 🔥 try it out. More to come soon......... 👀

Karan Goel3 months ago

👀

Daniele Paliotta3 months ago

That's technically a Pan Suisse but other than that I'm so so proud of team ❤️

Karan Goel3 months ago

🤌🤌🤌

Ricardo Buitrago3 months ago

is there a money back guarantee if my Cartesia agent can't upgrade me to business?

Brandon Chen3 months ago

i'll be flying with air cartesia from now on

Ronald Yu3 months ago

Z

Karan Goel3 months ago

A

Israel (Iz) Shalom3 months ago

no way you're making a 7:40am flight 😴

Stefano Ermon3 months ago

Congrats! Really impressive to see the team keep shipping frontier voice models at this pace 🚀🚀

Arjun Desai3 months ago

so much more coming

Kaitee3 months ago

Customer support has been running on Internet Explorer for years

Parul Gautam3 months ago

I actually tried calling the agent number in the video. it was crazy - could not tell it was AI. Almost makes me wonder if a real human is picking these calls and answering them.

Dipti Sharma3 months ago

why's the founder asking for business class?? broke behavior 💀

David W. Romero3 months ago

So proud of the team! <3 I do feel for @debblimm's croissant though😆

Vapi3 months ago

Exciting news for developers building voice experiences!

Hassan3 months ago

Congrats on the launch Karan and team! Looks great

madergaser3 months ago

Does air Cartesia have a loyalty program?

Karan Goel3 months ago

sign up and find out 😄

Addison Leong3 months ago

Ever since I was born, this is the only thing I've ever wanted.

Lucy Liu3 months ago

Can’t wait for airline support to be seamless

Rony3 months ago

Amazing how many Indian people will be able to interact with technology now without having to learn English

vamshi chowdary3 months ago

For those who dont want to type the URL: And if you start using Ink-2, you never have to 😉

Mira3 months ago

I genuinely wasn’t expecting the text-to-speech model to be this fast. I cloned my voice in the playground and the response speed honestly shocked me.

Taaruk3 months ago

Going from "AI that talks" to "AI that can actually have a conversation" is the shift. Congrats on Sonic 3.5 and Ink 2

Related Videos

Today we released Meta Spirit LM — our first open source multimodal language model that freely mixes text and speech. Many existing AI voice experiences today use ASR to techniques to process speech before synthesizing with an LLM to generate text — but these approaches compromise the expressive aspects of speech. Using phonetic, pitch and tone tokens, Spirit LM models can overcome these limitations for both inputs and outputs to generate more natural sounding speech while also learning new tasks across ASR, TTS and speech classification. We hope that sharing this work will enable the research community to further new approaches for text and speech integration.

Today we released Meta Spirit LM — our first open source multimodal language model that freely mixes text and speech. Many existing AI voice experiences today use ASR to techniques to process speech before synthesizing with an LLM to generate text — but these approaches compromise the expressive aspects of speech. Using phonetic, pitch and tone tokens, Spirit LM models can overcome these limitations for both inputs and outputs to generate more natural sounding speech while also learning new tasks across ASR, TTS and speech classification. We hope that sharing this work will enable the research community to further new approaches for text and speech integration.

AI at Meta

351,934 views • 1 year ago

There is a subtle architecture shift happening in voice AI. The voice stack is becoming part of the agent's execution loop. Cartesia is combining the listening and speaking paths around that loop. Sonic-3.6 turns text into speech (90ms latency) and Ink-2 turns speech into text (100ms transcript latency), faster than anything else streaming. now holds the #1 spot for both speaking and listening models. Becasue, voice agents are one of those systems where 100ms in the wrong place is very noticeable. Cartesia is attacking both sides of that loop at once. super interesting work by Karan Goel and team.

There is a subtle architecture shift happening in voice AI. The voice stack is becoming part of the agent's execution loop. Cartesia is combining the listening and speaking paths around that loop. Sonic-3.6 turns text into speech (90ms latency) and Ink-2 turns speech into text (100ms transcript latency), faster than anything else streaming. now holds the #1 spot for both speaking and listening models. Becasue, voice agents are one of those systems where 100ms in the wrong place is very noticeable. Cartesia is attacking both sides of that loop at once. super interesting work by Karan Goel and team.

Rohan Paul

310,254 views • 22 days ago

A startup with under 100 people just passed ElevenLabs, Gemini, and OpenAI on the voice leaderboards Cartesia took #1 on both speaking and listening in the same week. Sonic 3.5 for voice, Ink 2 for transcription👇

A startup with under 100 people just passed ElevenLabs, Gemini, and OpenAI on the voice leaderboards Cartesia took #1 on both speaking and listening in the same week. Sonic 3.5 for voice, Ink 2 for transcription👇

Nelly;

22,875 views • 3 months ago

MyRobot Dev Blog 1/14/26 - Quality of Life I added offline text to speech support (Chatterbox) so you can host your own offline text to speech models.

MyRobot Dev Blog 1/14/26 - Quality of Life I added offline text to speech support (Chatterbox) so you can host your own offline text to speech models.

sgthale | Family Time Dev

46,510 views • 6 months ago

Google presents AudioPaLM: A Large Language Model That Can Speak and Listen paper page: introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation. AudioPaLM inherits the capability to preserve paralinguistic information such as speaker identity and intonation from AudioLM and the linguistic knowledge present only in text large language models such as PaLM-2. We demonstrate that initializing AudioPaLM with the weights of a text-only large language model improves speech processing, successfully leveraging the larger quantity of text training data used in pretraining to assist with the speech tasks. The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training. AudioPaLM also demonstrates features of audio language models, such as transferring a voice across languages based on a short spoken prompt.

Google presents AudioPaLM: A Large Language Model That Can Speak and Listen paper page: introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation. AudioPaLM inherits the capability to preserve paralinguistic information such as speaker identity and intonation from AudioLM and the linguistic knowledge present only in text large language models such as PaLM-2. We demonstrate that initializing AudioPaLM with the weights of a text-only large language model improves speech processing, successfully leveraging the larger quantity of text training data used in pretraining to assist with the speech tasks. The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training. AudioPaLM also demonstrates features of audio language models, such as transferring a voice across languages based on a short spoken prompt.

AK

290,517 views • 3 years ago

We pioneered the first ultra-realistic Text to Speech model, and recently launched the world's most accurate Speech to Text model, Scribe. But we're not stopping there. Today, we're taking one small step for man, and one giant leap for man's best friend... with Text to Bark.

We pioneered the first ultra-realistic Text to Speech model, and recently launched the world's most accurate Speech to Text model, Scribe. But we're not stopping there. Today, we're taking one small step for man, and one giant leap for man's best friend... with Text to Bark.

ElevenLabs

291,881 views • 1 year ago

New text to speech models, they have range, pacing, accent, emotional expression. I gave Gemini 3 the launch blog post and asked it to make a short ad for these new models, but in an ASMR whisper:

New text to speech models, they have range, pacing, accent, emotional expression. I gave Gemini 3 the launch blog post and asked it to make a short ad for these new models, but in an ASMR whisper:

fofr

109,025 views • 9 months ago

Realtime speech-to-text for developers building conversational apps, AI agents, and voice workflows.

Realtime speech-to-text for developers building conversational apps, AI agents, and voice workflows.

ElevenLabs Developers

14,607,948 views • 4 months ago

is the best free speech to text tool and it just got better handy now supports streaming, and transcriptions are even faster now thanks to transcribe.cpp give v0.9.0 a spin and try the new models out

is the best free speech to text tool and it just got better handy now supports streaming, and transcriptions are even faster now thanks to transcribe.cpp give v0.9.0 a spin and try the new models out

cj

24,073 views • 3 months ago

I built a service desk agent using ElevenLabs’ new Conversational AI Agents feature. Watch the video to see how responsive it is! Previously, I used ElevenLabs for cloning my voice and used my generated voice for narrations in some of my youtube videos. This feature takes elevenlabs' voice AI to a whole new level! It simplifies systems that used to require separate TTS (text-to-speech) and STT (speech-to-text) processes for both sides of the conversation. Now, it’s much simpler! Create your own agent here What will you create with this?

I built a service desk agent using ElevenLabs’ new Conversational AI Agents feature. Watch the video to see how responsive it is! Previously, I used ElevenLabs for cloning my voice and used my generated voice for narrations in some of my youtube videos. This feature takes elevenlabs' voice AI to a whole new level! It simplifies systems that used to require separate TTS (text-to-speech) and STT (speech-to-text) processes for both sides of the conversation. Now, it’s much simpler! Create your own agent here What will you create with this?

Melvin Vivas

27,722 views • 1 year ago

BUILD 🔥: Microsoft is preparing new image and voice models for the announcement on June 2. > MAI Voice 2, a multilingual model supporting 15 news languages and a wider range of emotional spectrum (check voice samples in the article) > MAI Transcribe 1.5, a new model for speech-to-text use cases. > MAI Image 2.5, already announced last week, is now available on LM Arena in preview. Compared to MAI Image 2, it supports file uploads and can be used for image editing.

BUILD 🔥: Microsoft is preparing new image and voice models for the announcement on June 2. > MAI Voice 2, a multilingual model supporting 15 news languages and a wider range of emotional spectrum (check voice samples in the article) > MAI Transcribe 1.5, a new model for speech-to-text use cases. > MAI Image 2.5, already announced last week, is now available on LM Arena in preview. Compared to MAI Image 2, it supports file uploads and can be used for image editing.

🚨 AI News | TestingCatalog

46,535 views • 4 months ago

Day 1 of 3 days of MLX: Introducing MLX-Audio-Swift SDK 🚀 A modular Swift SDK for voice agents and tasks on Apple Silicon built by Lucas Newman and yours truly. iOS, macOS, and visionOS developers can now build native apps with real-time, on-device audio intelligence: 🗣️ Text-to-Speech (TTS) 👂 Speech-to-Text (STT) 🔄 Speech-to-Speech (STS) 🎙️ Voice Activity Detection (VAD) and more. Only import the capabilities you need, nothing extra. Get started today and leave us a star ⭐️

Day 1 of 3 days of MLX: Introducing MLX-Audio-Swift SDK 🚀 A modular Swift SDK for voice agents and tasks on Apple Silicon built by Lucas Newman and yours truly. iOS, macOS, and visionOS developers can now build native apps with real-time, on-device audio intelligence: 🗣️ Text-to-Speech (TTS) 👂 Speech-to-Text (STT) 🔄 Speech-to-Speech (STS) 🎙️ Voice Activity Detection (VAD) and more. Only import the capabilities you need, nothing extra. Get started today and leave us a star ⭐️

Prince Canuma

164,026 views • 7 months ago

🎙️Do you know you now have all the building blocks for full speech-to-speech? - Voxtral Realtime: High-quality, real-time speech-to-text. - Mistral Small 4: Fast, efficient, general-purpose agentic model. - Voxtral TTS: Realistic customizable text-to-speech with streaming output.

🎙️Do you know you now have all the building blocks for full speech-to-speech? - Voxtral Realtime: High-quality, real-time speech-to-text. - Mistral Small 4: Fast, efficient, general-purpose agentic model. - Voxtral TTS: Realistic customizable text-to-speech with streaming output.

Mistral for Developers

28,032 views • 6 months ago

introducing Gemini 3.5 Transcribe: our most precise speech-to-text model yet, designed for intelligent voice interactions this new model: - seamlessly handles self-corrections - removes filler words to deliver clean, formatted text - understands your natural intent and speaking style - captures audio, even in noisy environments - supports 85+ languages, including regional accents and diverse dialects try it today in the Gemini API and in AI Studio

introducing Gemini 3.5 Transcribe: our most precise speech-to-text model yet, designed for intelligent voice interactions this new model: - seamlessly handles self-corrections - removes filler words to deliver clean, formatted text - understands your natural intent and speaking style - captures audio, even in noisy environments - supports 85+ languages, including regional accents and diverse dialects try it today in the Gemini API and in AI Studio

Google AI Studio

238,290 views • 1 month ago

Speak with ChatGPT: You can now use voice to engage in a back-and-forth conversation with your assistant. The hyper-realistic text-to-speech model allows you to choose from five different voices. On mobile, opt-in to voice in Settings → New Features on the mobile app.

Speak with ChatGPT: You can now use voice to engage in a back-and-forth conversation with your assistant. The hyper-realistic text-to-speech model allows you to choose from five different voices. On mobile, opt-in to voice in Settings → New Features on the mobile app.

Rowan Cheung

55,915 views • 3 years ago

Mati Staniszewski on why ElevenLabs is betting on a cascaded approach for voice agents: “Our approach, as you think about voice agents, conversational agents, is effectively a cascaded approach. You use transcription or speech-to-text, an LLM, text-to-speech, and orchestrate all of that together. Then you have speech-to-speech, which goes directly from speech, and there’s a speech response on the other side. Today, we are optimizing heavily on a cascaded approach. As we work with a lot of the businesses and enterprises, they will need that visibility into what happens. They will want to execute certain tasks on top of that. They want good visibility into each of the steps and great accuracy of all the models. But beyond that, they can abstract away what’s the LLM layer, what’s the intelligence layer, and the integrations are easier in that system. That’s where we are betting a lot of the research work on how you can make that great, and we think we can make that great.” John Collison @matiii ElevenLabs

Mati Staniszewski on why ElevenLabs is betting on a cascaded approach for voice agents: “Our approach, as you think about voice agents, conversational agents, is effectively a cascaded approach. You use transcription or speech-to-text, an LLM, text-to-speech, and orchestrate all of that together. Then you have speech-to-speech, which goes directly from speech, and there’s a speech response on the other side. Today, we are optimizing heavily on a cascaded approach. As we work with a lot of the businesses and enterprises, they will need that visibility into what happens. They will want to execute certain tasks on top of that. They want good visibility into each of the steps and great accuracy of all the models. But beyond that, they can abstract away what’s the LLM layer, what’s the intelligence layer, and the integrations are easier in that system. That’s where we are betting a lot of the research work on how you can make that great, and we think we can make that great.” John Collison @matiii ElevenLabs

Stripe

18,605 views • 5 months ago

How to have your Openclaw install the new Qwen 3.5-9b Local LLM in 1 minute This came out TODAY and is one of the only local models you can run on a 16GB Mac Mini Will be testing and making some videos this week, look out for that (check below for some tests for now)

How to have your Openclaw install the new Qwen 3.5-9b Local LLM in 1 minute This came out TODAY and is one of the only local models you can run on a 16GB Mac Mini Will be testing and making some videos this week, look out for that (check below for some tests for now)

ashen

64,176 views • 7 months ago

Raon-OpenTTS-1B is a new open text-to-speech model for zero-shot voice cloning. It has best overall similarity score and lowest error rate vs other leading TTS models.

Raon-OpenTTS-1B is a new open text-to-speech model for zero-shot voice cloning. It has best overall similarity score and lowest error rate vs other leading TTS models.

AI Search

22,657 views • 1 month ago

Say It — private, local text-to-speech for macOS Open models 100% on-device, speak text from any app with a hotkey, and even clone your own voice. Infinite voices. Free & open-source. Listen to the video 🔊

Say It — private, local text-to-speech for macOS Open models 100% on-device, speak text from any app with a hotkey, and even clone your own voice. Infinite voices. Free & open-source. Listen to the video 🔊

calle

71,229 views • 1 month ago

From text to reality: MIT researchers find new ways to use LLMs to help in design & manufacturing. These models can convert text prompts to CAD, generate manufacturing instructions, and search for optimal designs:

From text to reality: MIT researchers find new ways to use LLMs to help in design & manufacturing. These models can convert text prompts to CAD, generate manufacturing instructions, and search for optimal designs:

MIT CSAIL

59,161 views • 3 years ago