Understand this story
On October 30, 2026, Shobhit Banga released Joy, a dual-track dialogue dataset containing 100,000 hours of real, unscripted two-person conversations, with each person on a separate audio track, covering 50,000 speakers and 10 languages. In a report the same day, Rohan Paul noted that current speech AI is mainly trained on fake conversations written by LLMs and read aloud by TTS, in which people cannot be heard interrupting each other, and he was curious about how these conversations are sustained for 10 to 20 minutes. The dataset has now been released, with no subsequent development progress yet.
AI-generated from public reports
Latest development · Report
Joy releases a 100,000-hour dual-track real conversation dataset covering 10 languagesDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 30
Rohan Paul
Joy releases a 100,000-hour dual-track real conversation dataset covering 10 languages
Shobhit Banga releases Joy, a dual-track conversation dataset containing 100,000 hours of real, unscripted two-person conversations, with each person on a separate audio track, covering 50,000 speakers and 10 languages. Rohan Paul points out that current speech AI is mainly trained on fake conversations written by LLMs and read aloud by TTS, which cannot capture people talking over each other, and he is curious about how these conversations are sustained for 10 to 20 minutes.