# Joy releases a dual-channel conversational dataset for voice models

> Original title: Nice to see conversation data that isn't decades-old telephone audio.

- Source: Elvis Saravia
- Published: 2026-10-29T18:25:04.000Z
- Bad Influence: https://badinfluence.news/items/cuto19wc6uhft7i57rqktml5t
- Original: https://x.com/omarsar0/status/2115872561116688459

## Summary

Joy is a dual-channel dataset of unscripted conversations between two people, each recorded on their own track, built to teach voice models turn-taking. It contains 100,000 hours of audio from 50,000 speakers across 10 languages. Elvis Saravia praises the release as the first dataset built around conversational timing rather than decades-old telephone audio.

## Post

Nice to see conversation data that isn't decades-old telephone audio.

Conversation is mostly timing. This is the first dataset release I've seen that's really built around that.

The 'this is only the start' part of the thread on @voicearena_ai's Joy caught my eye. 10x and 100x of this would be a lot. Really good work from this team.

## Quoted post @shobhitbanga

> Drop 1/15: Introducing Joy, a dual-channel conversational dataset.
> Real, unscripted conversations between two people, each on their own track. Built to teach voice models how people actually talk: when to listen, when to jump in, when to say “hmm”.
> 
> 100,000 hours. 
> 50,000 speakers. 
> 10 languages.

https://x.com/shobhitbanga/status/2115862853852995907