# Auto-ACD
Auto-ACD is a large-scale, high-quality, audio-language dataset, building on the prior of robust audio-visual correspondence in existing video datasets, VGGSound and AudioSet.


- **Homepage:** https://auto-acd.github.io/
- **Paper:** https://huggingface.co/papers/2309.11500
- **Github:** https://github.com/LoieSun/Auto-ACD

## Analysis

![](src/analysis.png)

Auto-ACD</strong>, comprising over <strong>1.9M </strong> audio-text pairs. 
As shown in figure, The text descriptions in Auto-ACD contain <strong>long texts (18 words)</strong> and <strong>diverse vocabularies (23K)</strong>, and provide information about the <strong>surrounding auditory environment</strong>(data point with <strong>shadow</strong>) in which sounds take place.

## Download

We provide a csv file. For each data pairs, we provide YouTube URLs and generated caption. Each line in the csv file has columns defined by here.

```
# YouTube ID, caption
```

## Dataset Preview

![](src/samples.png)