Multi-speaker conversation datasets with a separate recording for each speaker.

otoSpeech has English conversation among three or more people, also called multi-party or group conversation. Each speaker has their own recording, so overlapping speech is not mixed into one track.

otoSpeech 3-Speaker has three people talking, with one mono file per speaker. otoSpeech Face-to-Face has two to five people in one room, recorded on 16 channels at 48 kHz and 24-bit. Because they share a room, each microphone also picks up some of the other voices.

CasualLanguageSpeakersMediaLicenseActions
English2-5AudioVideoOn request
Request
English3AudioOn request
Request
Task-oriented
English2-4AudioVideoOn request
Request
Need larger groups or other languages? We record and annotate to your specification.Request

FAQ

What is multi-party conversation data?

Recordings of three or more people talking together. With more than two people, anyone can take the next turn, and two conversations can run at the same time.

How many speakers do these datasets have?

otoSpeech 3-Speaker has three. otoSpeech Face-to-Face has two to five, and its samples include three speakers, four speakers, and two pairs talking at once. otoSpeech Egocentric has two to four, with people working on a shared task.

How is this different from diarization?

Diarization estimates who spoke when from a mixed recording, and it struggles most during overlapping speech. Here each speaker has their own recording, so you start from the source.

Which languages are available?

otoSpeech 3-Speaker and otoSpeech Face-to-Face are in English. Tell us if you need other languages or larger groups.

Read more: About otoSpeech 3-Speaker and About otoSpeech Face-to-Face.

More otoSpeech data: English, Japanese, Korean, video, and egocentric conversation, or the full catalog.