---
dataset_info:
  features:
  - name: audiocap_id
    dtype: int64
  - name: youtube_id
    dtype: string
  - name: start_time
    dtype: int64
  - name: caption
    dtype: string
  splits:
  - name: train
    num_bytes: 4162928
    num_examples: 49838
  - name: validation
    num_bytes: 198563
    num_examples: 2475
  - name: test
    num_bytes: 454652
    num_examples: 4875
  download_size: 2781679
  dataset_size: 4816143
license: mit
task_categories:
- text-to-speech
language:
- en
multilinguality:
- monolingual
tags:
- youtube
- captions
pretty_name: AudioCaps
size_categories:
- 10K<n<100K
source_datasets:
- original
paperswithcode_id: audiocaps
---

# audiocaps

## Dataset Description

- **Homepage:** https://audiocaps.github.io/
- **Repository:** https://github.com/cdjkim/audiocaps
- **Paper:** [AudioCaps: Generating Captions for Audios in The Wild](https://aclanthology.org/N19-1011.pdf)

HuggingFace mirror of [official data repo](https://github.com/cdjkim/audiocaps).