---
# dataset card
# https://github.com/huggingface/hub-docs/blob/main/datasetcard.md?plain=1

language:
  - en
tags:
  - audio
  - speech
  - speaker-identification
  - speaker-verification
  - text-to-speech
  - voice-cloning
  - voxceleb
#task_categories:
# see list https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/src/pipelines.ts
#- text-to-speech
#- speaker-identification
license: cc-by-sa-4.0
pretty_name: VoxCeleb1 Speaker Voices & Embeddings
configs:
- config_name: default
  data_files:
  - split: train
    path: "audio/*.wav"

---
# VoxCeleb1 voice samples for SDialog

## Dataset Description

This dataset provides a collection of speaker voice samples from the [VoxCeleb1 dataset](https://mm.kaist.ac.kr/datasets/voxceleb/).
It is a subset of the VoxCeleb1 dataset, containing only the audio samples (~30s) and metadata.

The audio data is stored as mono-channel, 16000hz, signed 16-bit (little-endian) PCM wav files. 

### Dataset Components

1.  **Audio Clips**: 30-second `.wav` files for each speaker.
2.  **Metadata**: A `metadata.csv` file linking audio files to speaker information (ID, gender, name, etc.).


## How to use

### Loading the Dataset

The main dataset, containing the audio and metadata, can be loaded using the `datasets` library.

```python
from datasets import load_dataset

hf_dataset = "sdialog/voices-voxceleb1" 
dataset = load_dataset(hf_dataset)["train"]

# Access a sample
print(dataset[0])
```

## License

The VoxCeleb metadata is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. 
The audio data is scraped from publicly available YouTube videos. Data is only made available for academic purposes.