Skip to content

Dataset · HKUSTAudio

Multitask audio dataset for instruction-tuning

HKUSTAudio/Audio-FLAN-Dataset

Massive Audio-FLAN corpus in English and Chinese for speech, music, and sound understanding and generation.

What's inside

A large-scale dataset for instruction-tuning audio models (ranging from 10 to 100 million records in English and Chinese). It contains audio files and jsonl metadata categorized into speech, music, and general sounds (such as urban scenes from TAU and AudioSet data). The authors note that full file uploads are still ongoing.

What you can do with it

  • Training a unified audio model for automatic speech recognition (ASR) and text-to-speech (TTS)
  • Training sound effect and music generation systems from text descriptions
  • Conducting instruction-tuning of a multimodal model for understanding all audio types
Tasks
Text → speech, Text → audio, Automatic speech recognition
Languages
English, Chinese
Size
10M–100M rows
Downloads this month
863
Likes
47
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("HKUSTAudio/Audio-FLAN-Dataset")