Skip to content

Dataset · nvidia

Audio QA for expert reasoning

nvidia/AudioSkills

About 10 million audio-question-answer pairs for training audio-language models.

What's inside

AudioSkills-XL is a large audio-question-answer (AQA) dataset for developing audio-language models on expert reasoning tasks. It expands the original AudioSkills by approximately 4.5 million new QA pairs, totaling around 10 million examples. Audio fragments are short, up to 30 seconds; the language is English.

What you can do with it

  • Training audio-language models for questions on short audio
  • LoRA fine-tuning for expert audio QA
  • Evaluating audio reasoning in multimodal models
Tasks
Audio text → text
Languages
English
Size
1M–10M rows
Downloads this month
1.7K
Likes
107

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/AudioSkills")