Dataset · MultiTalk
Corpus for fine-tuning full-duplex voice dialogues
MultiTalk/MultiTalkFT
From 10K to 100K audio dialogues in English and Chinese for training conversational models with multiple speakers.
What's inside
A dataset for fine-tuning multi-party full-duplex voice communication models in English and Chinese (between 10,000 and 100,000 records). Metadata in jsonl files specifies paths to audio files and speaker voice prompts, clip durations in seconds, and system prompts with persona descriptions. Dialogue transcriptions are stored separately in parquet format.
What you can do with it
- Fine-tuning voice models for natural full-duplex dialogues
- Training speech synthesis and recognition systems with role and persona awareness
- Tasks
- Audio → audio
- Languages
- Chinese, English
- Size
- 10K–100K rows
- Downloads this month
- 27.1K
- Likes
- 11
How to load it
from datasets import load_dataset
ds = load_dataset("MultiTalk/MultiTalkFT")