# Voice Arena releases Joy, a dual-channel conversational speech dataset

> Original title: Really cool voice dataset from @voicearena_ai for training conversational models!

- Source: Santiago
- Published: 2026-10-29T18:05:59.000Z
- Bad Influence: https://badinfluence.news/items/mr2hjwwajdtwi6vr5ko5czbg2
- Original: https://x.com/svpino/status/2115867760941281621

## Summary

Voice Arena has released Joy, a dataset of real, unscripted conversations with each speaker on a separate audio track. The data covers 100,000 hours, 50,000 speakers and 10 languages, and includes overlapping speech, interruptions, pauses and backchannels such as "mm-hmm." Santiago says separate per-speaker tracks are necessary for training conversational models to get timing right.

## Post

Really cool voice dataset from @voicearena_ai for training conversational models!

This is for anyone who's researching or developing voice models.

The data here is pretty cool:

• Each speaker has a separate audio track
• You get access to overlapping speech
• Data includes interruptions, pauses, and "mm-hmm's"

If you want your models to get conversational timing right, you must use data with separate audio tracks for each speaker.

## 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