ILSpeech

Hebrew phonemes TTS Dataset 🗣️

Dataset content

This dataset contains around 2 hours of studio-quality speech.

It includes two speakers (merged into a single dataset):

• Speaker 1: 40m male speaker (~578 samples)
• Speaker 2: 1h 30m male speaker (~951 samples)

It includes clean Hebrew text and IPA phonemes. The audio is in 22.05 kHz mono format with clean recordings.

License

Non commercial. See LICENSE

Sample

• Speaker 1:

• Speaker 2:

Format

Annotations are provided in metadata.csv, metadata_train.csv, and metadata_test.csv (150 samples held-out test set).

Format: id|ipa|text — example: speaker1_000|kotˈeʁet|כותרת where speaker1_000 points to ./wav/speaker1_000.wav. Each wav is 2 to 25 seconds.

and the phonemization was done using https://github.com/thewh1teagle/phonikud, with manual corrections for accuracy.

Training

You can find training receipes for the dataset in github.com/thewh1teagle/ilspeech-train

Phonemize

The dataset comes with phonemes already so you don't need this for train,

However you do need it after you train the model.

you can phonemize the text using Phonikud

Citation

If you use this dataset in your work, please cite:

@misc{kolani2025phonikud,
  title={Phonikud: Hebrew Grapheme-to-Phoneme Conversion for Real-Time Text-to-Speech},
  author={Yakov Kolani and Maxim Melichov and Cobi Calev and Morris Alper},
  year={2025},
  eprint={2506.12311},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2506.12311},
}

Please note that the paper uses v1.0 of this dataset.