Listen to samples
The player shows the room in 3D with the seats in use lit up. Seat positions are approximate, and you can drag the view to rotate it. The last clip has two pairs talking at the same time.
- Two speakers
- Three speakers
- Four speakers
- Two pairs at once
The room
The room has 23.1 m² of floor area, a volume of 46.27 m³, and a 2.3 m ceiling. The four speaker stations sit in alternating positions, so neighbors and diagonal partners reach each microphone from different angles and distances.
The microphones
- Every station: a Sony C-38B and a Sony lavalier (ECM-77B or ECM-44B)
- Stations 1 to 3: a Shure SM7B and a Sennheiser MKE600
- Station 3: a Sennheiser MKH416
- Near the center of the room: an AKG C414 XLII
This is the main setup. Some later sessions used different microphones.
What the microphones hear
Everyone shares one room, so each close microphone also picks up the other voices. If you need clean isolation between speakers, otoSpeech 3-Speaker and our remote two-speaker sets are the better fit.
In-room work needs that bleed. Speaker separation, diarization, and far-field speech all have to deal with it, and here you can compare a lavalier, a studio microphone, and a room microphone on the same speech.
A humanoid’s microphones will sit in rooms like this one, hearing several people at different distances. Research on multi-party social robots lists overlapping speech among the hard problems.[1] This data does not include a robot’s own motor noise or speaker echo.
Data for voice AI
Get the dataset
Tell us what you are building and we will follow up on licensing. If you need larger groups, other rooms, or other languages, we can record to your specification.
References
- He Zhu, Ryo Miyoshi, and Yuki Okafuji. “Whom to Respond To? A Transformer-Based Model for Multi-Party Social Robot Interaction.” 2025. arxiv.org/abs/2507.10960