Dataset · zhifeixie
Streaming audio dialogues in English and Chinese (2.28M clips)
zhifeixie/StreamAudio-2M
A large multi-task streaming audio dataset for training audio LLMs and voice agents.
What's inside
The dataset contains about 2.28 million unique English and Chinese audio clips organized into streams (sequences of audio utterances in a unified scheme) across six subtasks. It includes subsets for audio understanding (90,738 edits based on AudioSet and FMA with descriptions and Q&A) and streaming ASR (28,109 streams based on CommonVoice and GigaSpeech).
What you can do with it
- Training multimodal audio LLMs and streaming voice agents
- Real-time ASR and sound scene recognition training
- Tasks
- Audio classification, Automatic speech recognition, Translation, Audio → audio
- Languages
- English, Chinese
- Size
- 100K–1M rows
- Downloads this month
- 1.4K
- Likes
- 32
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("zhifeixie/StreamAudio-2M")