Corpus description

Manually transcribed and speech-to-text aligned Galician ASR corpus containing 50 hours of multi-domain speech. The corpus contains different types of audios: conferences, debates, speeches, and interviews. The corpus is divided into three partitions: train (80%), dev(10%) and test (10%). Each partition contains audio fragments in WAV format, aligned with their respective transcriptions. In the "/data" folder you can find the audio partitions, while the "/transcript" folder contains the TSV files with the information of each audio file. The file naming scheme of the audio files consists of an ID indicating the type of audio ("conf", "deba", "entr", "disc"), followed by the speaker ID and the start and end time of the audio fragment (e.g.: "conf_Casalderrey_0.000_5.737.wav").

The corpus is divided into three partitions (train, dev and test) taking into account the number of hours and speakers per partition, e.g., the dev partition contains 10% of the total audio time of the dataset, and these audios are from a number of different speakers equal to 10% of the total number of speakers in the dataset. The three partitions are balanced to ensure that they all contain the same percentage of male and female audios, and type of audios (conferences, debates, speeches, and interviews). Futhermore, the sets of speakers included in the different partitions (train, dev and test) do not overlap.

Information about the exact number of hours contained in each of the partitions in the dataset can be found in the table below:

Partition	No. of hours
train	40.68
dev	4.771
test	4.501
Total	49.952

NOTE: partitions are not exactly split into 80%-10%-10% due to the balance restrictions we mentioned before.

Funding and acknowledgements “The Nós project: Galician in the society and economy of Artificial Intelligence” is possible thanks to the funding resulting from the agreement 2021-CP080 between the Xunta de Galicia and the University of Santiago de Compostela, and thanks to the Investigo program, within the National Recovery, Transformation and Resilience Plan, within the framework of the European Recovery Fund (NextGenerationEU). We would like to thank the Corpus Oral Informatizado da Lingua Galega (CORILGA) project for their kind collaboration in providing the original data.

We are also grateful to María Pérez Lago and Carmen Rodríguez Blumer for their review of some of the timestamps in this dataset.

For more information, please go to https://nos.gal/ or contact the Nós project at [email protected].

Terms and Conditions

By accessing and using this dataset, you agree to comply with all applicable laws and ethical standards regarding the protection of individual rights. The dataset contains voice files, transcripts, and metadata, including participant identity information, provided solely for research and development purposes. Users are strictly prohibited from using the dataset in any way that infringes upon the rights, privacy, or dignity of any individual represented within it. Any misuse, including but not limited to attempts to engage in discriminatory, harmful, or unlawful activities, is expressly forbidden.

Citation information

The original version of this dataset, in STM and TRS formats and non-segmented audio, can be found in Zenodo.

If you use this version of the dataset, please cite as follows:

Vladu, A. I., Vázquez Abuín, M., González Corbelle, J., Fernández Rei, E., García Díaz, N., Vidal Miguéns, A., & Magariños, C. (2024). Nos_Transcrispeech-GL

URL: https://huggingface.co/datasets/proxectonos/Nos_Transcrispeech-GL

Datasets:

proxectonos
/

Nos_Transcrispeech-GL

This repository is publicly accessible, but you have to accept the Terms and Conditions to access its files and content.

Collection including proxectonos/Nos_Transcrispeech-GL

ASR Datasets