---
license: mit
task_categories:
- text-to-speech
- text-to-audio
language:
- pt
tags:
- code
pretty_name: fake_voices
size_categories:
- 1B<n<10B
---
---
# Dataset Card for Fake Voices

This dataset contains deepfakes in Brazilian Portuguese created with [XTTS model](https://arxiv.org/pdf/2406.04904).

## Dataset Details

The dataset was created using the XTTS model, which is a Text-to-Speech model pre-trained in several languages including Portuguese. 
In order to generate the mentioned deepfakes, the model was fed with recordings from the [CETUC Corpus](https://gitlab.com/fb-audio-corpora), 
made available by Fala Brasil Group. It contains speeches from 101 speakers, totaling 140 hours of audio.

### Dataset Description

<!-- - **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed] -->
- Language(s) (NLP): Brazilian Portuguese
- License: MIT

<!-- ### Dataset Sources [optional] -->

<!-- Provide the basic links for the dataset. -->

<!-- - **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed] -->

## Uses

This dataset was created for Portuguese deepfake detection purposes, enabling research in various areas, including audio classification, signal processing and statistical analysis. 

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

The dataset contains .zip files, each of them representing one speaker. In each .zip, there are deepfakes and their corresponding .txt transcriptions.

## Dataset Authors 

- Éric Carvalho Figueira
- Marcos Godinho Filho

## Dataset Contact

Contact us via unfake.official.contact@gmail.com.