Hebrish

About the project

Acknowledgements

Hebrish is a personal project, and much of what it knows comes from other people's work. These are the organisations and people who have given it data, recordings and help, and what each one contributed. Every source is also credited on the dataset it went into.

Data supporters

Hebrish thanks these organisations for their support. Names, streets and places are where speech recognition fails most often: a model that has never heard a name writes it wrong. These public bodies publish the lists the name and place sentences are built from, and their staff took the time to answer questions about them.

Central Bureau of Statistics הלשכה המרכזית לסטטיסטיקה

Contribution: the first names given to babies born in Israel from 1949 to 2024 (release 391/2025, Table 1: every name given to at least five babies). It decides which first names the Israeli Name Sentences are written for, most common first.

With thanks to Aviel Kranzler, head of the higher education and science statistics field and co-author, with Uri Naftalovich, of the first-names and surnames releases, for his answers about surname data.

data.gov.il, National Digital Agency מערך הדיגיטל הלאומי

Contribution: the national street list (63,662 streets in 1,316 localities) and the localities list behind Israeli Street Names; and the Population and Immigration Authority's list of first names in the Population Registry, cross-checked against the CBS birth counts to add 329 common first names that were still missing (2026-11-19).

With thanks to Tomer Sharon, DataGov product manager, who pointed the project to the first-names and multilingual localities datasets.

Jerusalem Institute for Policy Research מכון ירושלים למחקרי מדיניות

Contribution: the names of Jerusalem's statistical areas and sub-quarters in Hebrew and English, as used in the Statistical Yearbook of Jerusalem. They added 26 Jerusalem neighbourhoods the Israeli places task lacked (2026-11-19) and give an authoritative English spelling.

With thanks to Omer Yaniv, researcher, who prepared the list.

Recordings

The models are trained on Daniel's voice. These contributors have also recorded for the project (the speaker registry has their details):

  • Hannah: recordings in a US English accent, since 2026-10-29.
  • Mike: recordings in an Australian English accent.

More voices make the models better for everyone. Send recordings and your name goes here.

Open data and software

  • OpenStreetMap contributors: English street and neighbourhood names (ODbL).
  • Sefaria: the English Jewish texts read for the Judaica Sentences (CC0, CC-BY and public domain editions only).
  • LibriSpeech: replayed during training to keep the model's English.
  • ivrit.ai: the Hebrew streaming Zipformer the Hebrew speech-to-text work starts from.
  • The base models and the software the apps run on are listed under Credits on the About page.

Helped and not listed?

If you or your organisation contributed and are missing here, or want your entry changed, get in touch.