Dataset Card for QuEsT

The Quechua Spanish Translations dataset (QuEsT) is a language pair dataset of quechua sentences and its translations to spanish.

Dataset Details

Dataset Description

The QuEsT Dataset is conformed of two subsets: train and test. In this section we are going to refer as dataset to the train subset only.

The dataset consists of pairs of audios and their transcriptions in the languages of Quechua Cuzco (quz) and Spanish (spa), each pair being conformed by the translation of the sentences from one language to the other.

We were able to collect about 42 minutes of Quechua and 34 minutes of Spanish from 6 speakers, being 2 men and 4 women. Around 49 minutes (64% of the audio) were recorded from women, while around 27, (36%) from men.

• Language(s) (NLP): quechua, spanish
• License: GPLv3 due to inclusion of GPL-licensed materials

Uses

Direct Use

The aim of this dataset is to train text-to-speech models and machine translation models with quechua and/or spanish languages.

Dataset Structure

Quest

• train
  • speaker_name
    • transcripts.txt
    • language id (qu, es)
      • *audio_files.wav
• test
  • language id (qu, es)
    • *audio_files.wav

Dataset Creation

Source Data

Crowdsourced data of quechua speakers from Cusco.

Citation [optional]

BibTeX:

[More Information Needed]