Progress
How far Hebrish has come since the first recording: the running totals, and how each one has grown.
43.1hours of recorded speech to train on8,407code-switching recordings3,642Hebrew words and names in the lexicon2,778of them recorded at least once67published datasets34model repos on Hugging Face6,102model downloads147,373dataset downloads2voices so far27days since the first recording
Counted 2026-11-09 from the files listed under Sources. Hover a number for where it comes from.
Over time
Sources
- Hours: Daniel's recordings in the training split of the master dataset, counted at each model release; general English from public audiobooks and other voices are not counted (dataset growth).
- Recordings: the public code-switching dataset, train and test, by recording date; refreshed daily.
- Lexicon: the lexicon explorer; a word counts as recorded when a transcript contains it.
- Downloads: Hugging Face's own counters across every Hebrish model and dataset repo, one reading per day they were taken.
The same numbers as one file: progress.json.