README.md

metadata

license: cc-by-4.0

Dataset Card

This dataset is the aligned phoneme subset of the DoReCo South England dataset, split into utterances + phonetic transcriptions based on pause lengths / total length, with the goal of creating utterances < 30s for fine-tuning speech recognition models on phoneme recognition, not one phoneme at a time, but rather for entire utterances.

It is already randomly pre-split into train/dev/test sets, with 80% in train, 10% in dev, and the final 10% in test.

---

Link to original dataset website: https://doreco.huma-num.fr/languages/sout3282

Original dataset citation:

@incollection{doreco-sout3282,
  address   = {Berlin \& Lyon},
  author    = {Schiborr, Nils Norman},
  booktitle = {Language Documentation Reference Corpus (DoReCo) 1.2},
  editor    = {Seifart, Frank and Paschen, Ludger and Stave, Matthew},
  publisher = {Leibniz-Zentrum Allgemeine Sprachwissenschaft \& laboratoire Dynamique Du Langage (UMR5596, CNRS \& Université Lyon 2)},
  title     = {English (Southern England) DoReCo dataset},
  url       = {https://doreco.huma-num.fr/languages/sout3282},
  doi       = {10.34847/nkl.9c271u5g},
  urldate   = {16/01/2024},
  year      = {2022}
}