Meorav v2.2: the Hebrish voice model, on 5.5 hours of recordings
More than three times the training data of v2.1, a new enhancement model, and the same training settings.
Meorav v2.2 was released on 2026-11-09. Meorav is the Hebrish text-to-speech model: it reads English with romanized Hebrew words in it ("Hebrish") aloud in Daniel Rosehill's voice, with the Hebrew words said the Israeli way. It is a full fine-tune of Qwen3-TTS-12Hz-1.7B-Base. v2.2 replaces v2.1, and the main change is the training data: more than three times as many takes, cleaned with a better enhancement model.
Samples
Ten sentences, shortest first, each as a Before and After pair plus two more takes from Meorav, as they came out of the models; none was picked or edited. Before is the untuned base model, Qwen3-TTS-12Hz-1.7B-Base, cloning Daniel's voice zero-shot from one reference recording (the same reference take training used). After is Meorav v2.2 at the default sampling temperature (0.9). The last two players are the same Meorav weights at temperature 1.1 and 1.3: a higher temperature lets the model pick less likely prosody, which can sound livelier and less flat, at some risk of odd stresses. Short texts are generated in one pass; longer ones sentence by sentence, then joined.
Yalla, let's go, we're already late for the brit mila.
We're doing a quiet shabbat this week, just challah, some chamin and an early night.
I spent the whole morning at Misrad Hapnim, and I still don't have the teudat zehut.
Everyone here says everything will happen acharei hachagim. I used to find it frustrating, but honestly, after a few years you start saying it yourself.
On Friday morning I went down to the shuk for the usual run. The guy at the bourekas stall remembered me, gave me an extra one, and told me to say shabbat shalom to the kids.
The arnona bill came in twice this month, so I rang the iriya. After forty minutes on hold they told me, beseder, it's a mistake, just ignore the second one.
When I started ulpan, the morah told us not to worry about the grammar yet. Just listen, she said, and the words will come. She was right about most of it, although I still mix up my verb patterns when I'm tired. These days I can argue with the makolet guy about the price of cottage cheese, which feels like real progress.
Our shul does a kiddush after davening every shabbat, and it has become the real centre of the week. You hear about every simcha, every new oleh, and who is looking for a chavruta. The gabbai keeps a list of who is bringing what, and somehow there is always too much kugel. If you are new to the neighbourhood, honestly, just turn up, nobody will let you stand on your own for long.
Last chol hamoed we did a tiyul up in the Galil with the whole mishpacha. We stopped for shakshuka at a little place near the Kinneret, then walked along the Jordan for an hour. My father-in-law kept telling everyone it was nothing compared to the old days, before the road was paved. By the time we got back to Yerushalayim it was nearly midnight, and the kids were fast asleep in the back.
When we made aliyah, the first few months were a blur of paperwork. There was the misrad haklita, the kupat cholim, Bituach Leumi, and a bank that wanted three copies of everything. Our neighbour, a sabra who had seen it all before, sat us down with a cup of coffee and said, savlanut, it all gets done eventually. She was right. By our first Pesach in the country we had a lease, a car, and a teudat oleh that was already falling apart from use. Looking back now, the bureaucracy is not what I remember. What I remember is the people who helped, without being asked, and the way the whole street seemed to come out for Yom Haatzmaut.
To compare them with Daniel's real recordings and with the base model, see Hebrish/Meorav-Samples on Hugging Face.
Where to get it
Hebrish/Meoravon Hugging Face: the model forqwen-tts, voicedaniel. Access is by request, like the other Hebrish models.Hebrish/Meorav-GGUF: GGUF builds in Q8_0 and Q4_K_M for running on a device. The same files are on the Hebrish CDN for on-device use on Android.- Meorav (beta) for Android, 0.1.0: download the APK (2.99 MB). It runs the model on the phone with qwentts.cpp and downloads the v2.2 voice from the Hebrish CDN on first use. It is a beta, and Android only.
The model page lists approximate hardware requirements for generating speech.
The model page has example code and the version history.
What changed from v2.1
- More data. v2.1 was trained on 1,103 read takes, about 88 minutes. v2.2 is trained on 3,995 takes, about 5.5 hours.
- Takes from the speech recognition recordings. Most of the new takes (2,924) come from the recordings Daniel makes for Hebrish's speech recognition models. They were screened down to code-switching sentences only: no tech vocabulary, no digits and no noisy rooms. Sound-event tags from PANNs and quality scores from DNSMOS did the screening. Takes over 15 seconds were left out.
- A different enhancement model. Every take is cleaned with NVIDIA NemotronLabs RE-USE (
nvidia/NVIDIA-NemotronLabs-RE-USE, OpenMDW-1.1 licence), run as theaudio-cpp/RE-USE-GGUFbuild through audio.cpp. It replaces DeepFilterNet 3, which cleaned v2.1's audio. - The same training. After enhancement, the takes go through a podcast-style voice chain (EQ, de-esser, compression, loudness to -20 LUFS). Training uses v2.1's settings: 3 epochs at learning rate 7e-6.
How it's made on the model page lists every step and every model used.
The enhancement stage
Daniel's takes are recorded at home: real rooms, a crying baby now and then, a phone or desktop microphone. Before training, every take goes through NVIDIA NemotronLabs RE-USE, a speech-restoration model that removes background noise and room sound and keeps the voice. Then a podcast-style chain finishes it: EQ, de-essing, compression and loudness to -20 LUFS.
This matters because a text-to-speech model learns everything in its training audio, the noise included. Cleaner input means cleaner output.
Audio enhancement by NVIDIA NemotronLabs RE-USE. nvidia/NVIDIA-NemotronLabs-RE-USE (OpenMDW-1.1 licence), run through audio.cpp from the audio-cpp/RE-USE-GGUF build.
Choosing the enhancement model
DeepFilterNet 3 did this job for v2.1. For v2.2, two enhancement models were tried: RE-USE and Sidon v0.1. On one test take with a crying baby in the background, both removed the baby completely, and neither left the robotic tinge that older noise suppressors leave on a voice. By ear, Daniel gave RE-USE a slight edge. A spectrogram comparison decided it: RE-USE keeps the speaker's own detail above 4 kHz, where Sidon re-synthesises it.
Evaluation
There is no benchmark or listening score for Meorav. v2.2 was released after Daniel listened to it, as v2.1 was. The samples above let you judge it for yourself.
Earlier versions
Meorav v2.1 and v2.0 (both 2026-10-22) stay in the history of Hebrish/Meorav. The first training run, published as Hebrish/TTS-V1, is no longer on Hugging Face. The full list is on the model page and in the release notes.
