ymoslem/Tatoeba-Speech-Irish

Dataset Details

Synthetic audio dataset, created using Azure text-to-speech service. The bilingual text is a portion of the Tatoeba dataset, consisting of 1,983 text segments. The dataset consists of two sets of audio data, one with a female voice (OrlaNeural) and the other with a male voice (ColmNeural). The speech data comprises approximately 2 hours and 39 minutes (02:39:31) spread across 3,966 utterances.

Dataset Structure

Dataset({
    features: ['audio', 'text_ga', 'text_en'],
    num_rows: 3966
})

Copyright

Please refer to the Terms of Use of Tatoeba project.

Citations

• This speech dataset Tatoeba-Speech-Irish was created and introduced as part of the following paper:

@inproceedings{moslem2024leveraging,
  title={Leveraging Synthetic Audio Data for End-to-End Low-Resource Speech Translation},
  author={Moslem, Yasmin},
  booktitle={Proceedings of the 2024 International Conference on Spoken Language Translation (IWSLT 2024)},
  year={2024},
  month={April},
  url={https://arxiv.org/abs/2406.17363},
  address={Bangkok, Thailand}
}

• The original Tatoeba bilingual text dataset was collected by the OPUS project, introduced in the following paper:

@INPROCEEDINGS{Tiedemann2012-OPUS,
  title     = "{Parallel Data, Tools and Interfaces in {OPUS}}",
  booktitle = "{Proceedings of the Eighth International Conference on Language
               Resources and Evaluation ({LREC}'12)}",
  author    = "Tiedemann, J{\"o}rg",
  publisher = "European Language Resources Association (ELRA)",
  pages     = "2214--2218",
  month     =  may,
  year      =  2012,
  url       = "http://www.lrec-conf.org/proceedings/lrec2012/pdf/463_Paper.pdf",
  address   = "Istanbul, Turkey"
}