SmoothConv
100-hour Chinese multi-channel full-duplex conversational speech dataset with expert fine-grained annotations
Duration
100.53 hours
Languages
1
Sample Rate
48 kHz
Published
2026-06
Description
1Contains 100.53 hours of naturally occurring multi-party Chinese conversation recordings across tutoring and social chat scenarios
2Recorded in multi-channel environments and captures realistic conversational dynamics such as overlapping speech, backchannels, interruptions, pauses, and turn transitions
3Includes 2,503 audio files, with a mean duration of 144.59 seconds and an utterance duration range from 60.0 to 634.7 seconds
4The repository is organized into multi-channel subsets such as 2ch, 3ch, and 4ch; a sampled 2ch audio file was 48 kHz, 16-bit PCM
5Manually annotated by trained experts with human transcripts, speaker labels, turn states, emotion, age, gender, non-speech events, and scene/environment labels
6Supports turn labels including complete, incomplete, backchannel, and wait, and targets turn-taking modeling, overlap and interruption detection, full-duplex spoken dialogue systems, conversational speech understanding, and Speech LLM research
7Released under the CC BY-NC 4.0 license
Language Details
| Language | Duration |
|---|---|
| Chinese | 100.53 hours |
Publisher
License & Commercial Use
License
CC BY-NC 4.0
Commercial Use
Non-commercial only