Datasets:

BSC-LT
/

CAESAR-TINY

language:
  - ca
  - es
license: gpl-3.0
multilinguality:
  - code-switching
tags:
  - project aina
  - Barcelona Supercomputing Center
  - code-switching
task_categories:
  - automatic-speech-recognition
task_ids: []

Dataset Card for CAESAR-TINY

Dataset Description

Homepage: Project Aina
Repository: CAESAR-TINY

Dataset Summary

CAESAR-TINY is a synthetic code-switched dataset generated by combining monolingual samples in Catalan and Spanish. The process includes trimming silences, normalizing audio volume, and introducing random pauses. It contains 2 hours of speech data, created by concatenating audio from the Common voice 17 Benchmark split and VoxForge Spanish datasets.

Example Usage

To load CAESAR-TINY:

from datasets import load_dataset
caesar_tiny = load_dataset("BSC-LT/CAESAR-TINY", split="train")

Supported Tasks

The CAESAR-TINY dataset is designed for the Automatic Speech Recognition (ASR) task, enabling the transcription of utterances in Catalan, Spanish, and code-switched speech between the two languages.

Languages

The dataset features code-switched speech, combining Catalan (ca) and Spanish (es) within the same audio samples.

Dataset Structure

Data Instances

{
'audio': {
          'path': '14.wav',
          'array': array([0., 0., 0., ..., 0., 0., 0.]),
          'sampling_rate': 16000
          },
'transcription': 'trons del cul tempestat de merda los mismos ocultándose para volver a aparecer con regularidad casi mecánica'
}

Data Fields

audio (datasets.Audio) - a dictionary containing the path to the audio, the decoded audio array, and the sampling rate.
transcription (string) - normalized audio-segment transcription.

Data Splits

The dataset consists of a single split due to its limited size.

Dataset Creation

Curation Rationale

This corpus specifically focuses on Catalan-Spanish code-switched, a linguistic phenomenon that is very common in the daily lives of Catalonians. This task is particularly low-resourced because, besides being a variety of the Catalan language, it further restricts the available data by incorporating code-switching, a complex and less-explored aspect of language use. With this release, we develop the first Catalan-Spanish CS dataset, which will be valuable mainly for training and evaluating Code-Switching Speech Recognition systems in Catalan and Spanish.

Source Data

The dataset was created by concatenating original audio from the Common voice 17 Benchmark split and VoxForge Spanish datasets.

Data Collection and Processing

The dataset was created using a two-step pipeline for generating synthetic code-switched speech data from monolingual sources based on NeMo scripts. First, an intermediate manifest file was generated, which pairs utterances from two monolingual datasets based on specified language codes, duration constraints, and overall dataset size requirements. Next, we synthesized the speech data by concatenating selected segments, applying configurable pauses at the beginning, between segments, and at the end of each sample. The resulting dataset maintains linguistic diversity while ensuring consistency in audio normalization and sampling rate. This approach enables the creation of large-scale, high-quality code-switched datasets suitable for training and evaluating multilingual ASR models.

Annotations

The dataset doesn't contain any additional annotations.

Considerations for Using the Data

Social Impact of Dataset

CAESAR-TINY is a source of code-switching speech data that will be valuable in the development of speech technologies for Catalan and Spanish.

Discussion of Biases

No specific bias mitigation strategies were applied to this dataset. Inherent biases may exist within the data.

Other Known Limitations

Gender in the speech recordings is identified, but one or more speakers could be speaking in the same recording. For this reason, we don't know the total number of speakers in the corpus.

Dataset Curators

The corpus was curated by Abir Messaoudi during 2024 at the Barcelona Supercomputing Center.

Licensing Information

GNU General Public License v3.0

Citation Information

@misc{caesar-tiny-bsc2024,
      title={CAESAR collection for Catalan and Spanish Code-Switching datasets},
      author={Messaoudi, Abir and Solito, Sarah and Kulebi, Baybars},
      publisher={Barcelona Supercomputing Center},
      year={2024},
      url={https://huggingface.co/datasets/BSC-LT/CAESAR-TV3}
}

Contributions

This work has been promoted and financed by the Generalitat de Catalunya through the Aina project.