CLAUSE-Bielefeld/10M_phonemized_German_dataset

Small Phonemized German dataset

Description

This dataset is a smaller and phonemized version of Bastian Bunzeck's dataset (Bunzeck et al., 2025) and contains approximately 10 million words.

The phonemization was done using Phonemizer by Bernard & Titeux, 2021.

This dataset was used to train the Phoneme-based German babyLlama model.

---

Source Composition

The dataset has been compiled from the following sources:

Source	Description	Words
CHILDES	Child-directed speech	3,626,301
CHILDES	Child speech	31,511,144
Flluter	German youth magazine	2,862,278
Klexikon	Children's online encyclopedia	1,384,891
MiniKlexikon	Simplified online encyclopedia	272,886
Dreambank	Dream reports	939,197
Total	--------	10,596,697

---

References

@misc{bunzeck2025constructiondistributionsshapeformal,
      title={Do Construction Distributions Shape Formal Language Learning In German BabyLMs?}, 
      author={Bastian Bunzeck and Daniel Duran and Sina Zarrieß},
      year={2025},
      eprint={2503.11593},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.11593}, 
}

@article{Bernard2021,
  doi = {10.21105/joss.03958},
  url = {https://doi.org/10.21105/joss.03958},
  year = {2021},
  publisher = {The Open Journal},
  volume = {6},
  number = {68},
  pages = {3958},
  author = {Mathieu Bernard and Hadrien Titeux},
  title = {Phonemizer: Text to Phones Transcription for Multiple Languages in Python},
  journal = {Journal of Open Source Software}
}