Eleven v3 Conversational, our most expressive model for realtime speech, is now generally available. For developers building voice experiences that respond with real emotion, Eleven v3 Conversational includes audio tags for fine-grained control and support across 70+ languages.

Reliable, high quality, and expressive speech. Eleven v3 Conversational has been optimized specifically for realtime use cases, with consistent quality across streaming generations and a library of 11,000+ voices.

Audio tags give fine-grained control over emotion and delivery. Direct voices to laugh, whisper, act sarcastic, or express curiosity with a broad range of tags.

Global language support. v3 Conversational supports 70+ languages, excelling across German, Spanish, French, Portuguese, and Hindi.

Watch the full walkthrough by @ElevenLabsDevs

Available now on ElevenAgents and ElevenAPI. From $0.05/1k chars, decreasing with scale.

There is high demand for quality Kurdish (Sorani) sound generation. I hope you add Kurdish support soon.

saudi najdi dialect included ? if not im definitely not interested

You guys are getting too expensive for enterprise use at scale! Love the product but unusable

Realness is not a studio-quality voice. When I call somebody on a phone, I'm not getting that realness in the voice. The demo is good for maybe production work, but not support agents. The realness is not there. The quality is too high.

base v3 hit GA back in march and was explicitly built for pre-rendered audio, not real time, they always recommended flash for that. this "conversational" variant is the missing piece then, real time voice agents finally getting that emotional range instead of the flatter delivery flash was stuck with

Excited to try it!

keep glowing !!

Amazing!

yikes imagine being actually mad and some fake ass ai hits you with a "ohhh wowww I totalllllly understannnd your fustrationnn"

It's incredible how they ignore the requests for support. No one taking responsibility for thousands of hacked accounts. Wait till the formal complaint filing is completed with the card brands. This is blatant fraud on what was a great legit company. Do the right thing help us.

It's good thing

Eleven v3 Conversational is live now with real emotion

Supports 70+ languages and fine-grained emotional control. We are entering an era in which AI assistants will not only speak fluently but also convey tone, intent, and personality. We look forward to seeing what developers build with this capability.

Does this support Indian languages?

🔥

the honest answer is that we still haven't cracked the code on affective computing. eleven v3 conversational gets us closer, but there's still a lot of nuance to figuring out how humans really respond to emotion in voice interfaces

This is my last attempt to bring to your attention that hundreds of customer accounts are experiencing account hijacking and having their credit and se it cards compromised due to negligence on the part of your having changed your support team in favor of non human assistance.

Kind of insane.

Cost ??

Expressive realtime speech across 70 languages is a major leap

70+ languages with real-time expressive speech is huge. Voice AI is getting seriously lifelike.

i've seen audio tags hurt trust when the model misreads user emotion

Audio tags and 70+ languages are useful surface features. The realtime test is interruption recovery: can the model stop speaking, preserve the partial turn, and resume with the right prosody?

supporting 70+ languages makes the model useful for voice products serving global users

calling for ElevenLabs researchers, product managers & others responsible for voice agents/speech-to-text models worthy to take a look at recent humyn labs' article about the way to decrease the word error rate for cases when its used by bilinguals

Why don't you just record 1 hour of a random American-based customer service rep and emulate it? Or 999 call person? They have a relaxed, neutral and static temperament that does not sound robotic like your current stuff

it's a little too breathy for a customer service agent. maybe tone it down slightly.

Do the audio tags actually work the same in all 70? I put out the same newsletter in 8 languages every week and that's usually where stuff starts drifting. Haven't tried the tags myself yet, just plain text, so I'm genuinely asking.

monotone never made it onto anyone's timeline. a cheerful voice reading your fraud alert gets screenshotted within the week

Benchmarks?

Expressive realtime voice across 70 languages is a huge leap forward