Multi-speaker dialogue at scale
Podcasts, interviews, panels and roundtables with two or more speakers. Separate tracks per speaker preferred.
Data owners list consented conversation, meeting and interaction recordings. AI labs and researchers find them, license them and pay the owner. Crosstalk handles consent, rights and delivery.
Open requests from research labs and model builders. If you hold data like this, register it and we'll match you.
Podcasts, interviews, panels and roundtables with two or more speakers. Separate tracks per speaker preferred.
Recorded meetings, sales and support calls, plus email and documents, from startups that have closed. Sold with proper authority and PII handling.
Conversations in specialist settings such as healthcare. Simulated clinical encounters and health-focused shows are a good starting point.
Describe the language, domain, group size and hours, and we'll source it from owners in the network.
Licensable data from owners in the Crosstalk network, alongside the best public Social AI datasets, with license and consent status on every entry.
Any setting, group size, language or volume. We'll scope a batch from our library or run a custom collection for you.
Listing is free. You keep ownership, choose what to share, and get paid each time it's licensed.
Tell us what you have: type, language, hours and format. Add a sample if you like.
We confirm you own it and that every participant consented, or help you collect releases.
Your listing goes to labs with open requests that fit. You approve every license.
You earn a share of every license, paid out as buyers license your data.
After review we may ask for a short 2–5 minute sample by email. It is shared only with the Crosstalk team.
We'll review rights and consent and email within two business days. Once it's approved, it goes to buyers with matching requests.
What Social AI is, how it relates to physical AI, and why consented, multi-party interaction data is becoming the scarcest input in computational intelligence.
Read the paper →DeepMind tested an AI mediator with more than 5,000 people. The lesson is about what group conversation demands from machines.
2 min readRead more ConversationTurn-taking is one of the hardest problems in conversation, and a new generation of models is finally tackling it.
2 min readRead more Physical AIThe first commercial humanoid jobs are here. The most revealing detail is how far they are kept from people.
2 min readRead more
Social AI learns from every kind of human interaction.
Social AI understands how people interact: who speaks next, who they're looking at, when someone is interrupted, how a group moves. It learns from recordings of real people interacting, with every participant's consent.