Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
audio
audioduration (s)
0.55
511
End of preview. Expand in Data Studio

Hebrish Demos

Free, ungated demo audio for the Hebrish models: samples, voice references and longer pieces, so they can be shared, linked and embedded without logging in or accepting terms. The models themselves are gated; this repo holds only their outputs.

Layout: <task>/<model>/<version>/…, where task is tts (text to speech) or asr (speech recognition). index.json lists every entry in machine-readable form.

Path Task Model Version What it holds
tts/meorav-multi-voice/v1.0/ TTS Hebrish/Meorav-Multi-Voice v1.0 (2026-11-16) samples/<voice>/NN.mp3: the 10 lines of samples/lines.json in each of the five designed voices (seed 0, final checkpoint); voices/*.wav: the reference clips that select each voice; audio-drama/: The Teudat Zehut, an 8 min 31 s (510.9 s) comedy with all five voices and Daniel's
tts/meorav-daniel/v1.0/ TTS Hebrish/Meorav-Daniel v1.0 (= Meorav v2.2) samples/: release samples (-base = the untuned Qwen3-TTS base on the same text, -t1.1/-t1.3 = other sampling settings); long-form/: 15 s and 30 s passages, joined vs one-shot; texts in samples.json
tts/meorav/v3.0/ TTS Hebrish/Meorav v3.0 (2026-11-15) short samples, side-by-side/ comparisons, and The Voice from Nachlaot (story text in the .md)

The meorav-daniel and meorav files are copies of those in the Space Hebrish/Meorav-Samples, which plays them.

Voices

All Meorav-Multi-Voice and Meorav v3 voices are designed (generated from a text description), not clones of a real person. Meorav-Daniel is Daniel Rosehill's own voice.

Licence

The audio and texts here are CC BY 4.0: share and reuse with credit to Hebrish (hebrish.app). The models that made them have their own licences (CC BY-NC 4.0 plus the Hebrish terms).

Downloads last month
56

Space using Hebrish/Demos 1

Collections including Hebrish/Demos