Is this your dataset? Claim it with the Hugging Face account that owns it.

MultiTalk/
MultiTalkPT

Name: MultiTalkPT
Creator: MultiTalk
License: https://spdx.org/licenses/CC-BY-NC-4.0.html

Views

No views yet

MultiTalkPT

Pre-training corpus for full-duplex spoken-dialogue models.

Schemas

data_{zh,en}.jsonl (one record per line):

field type description

path string relative path to the dialogue audio

voice string relative path to the speaker prompt audio

duration float dialogue duration in seconds

system string persona / system prompt

transcripts/*.parquet:

column type description

audio_path string matches data_*.jsonl path

id string… See the full description on the dataset page: https://huggingface.co/datasets/MultiTalk/MultiTalkPT.