word string | heard_as list | kinds list | meaning string | occurrences int64 | models list | first_seen timestamp[s] | last_seen timestamp[s] | takes int64 | target int64 | status string |
|---|---|---|---|---|---|---|---|---|---|---|
ROCm | [
"ROCM",
"Rockham",
"Rockin",
"RCM"
] | [
"tech",
"linux"
] | AMD's open-source GPU compute stack (like CUDA) | 13 | [
"shakshuka Shakshuka v2.0"
] | 2026-10-21T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Hebrish | [
"HeBrish",
"Hebris",
"keybrush"
] | [
"personal"
] | Daniel's Hebrew-English code-switching ASR project, app and Hugging Face org | 3 | [
"sabich Sabich v2.0",
"shakshuka Shakshuka v2.0"
] | 2026-10-21T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Nemotron | [
"Nematron",
"Nementron"
] | [
"tech",
"ai_interactions"
] | NVIDIA's Nemotron model family (a streaming ASR model here) | 3 | [
"sabich Sabich v2.0",
"shakshuka Shakshuka v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
F-Droid | [
"FDroid"
] | [
"tech"
] | the open-source Android app store | 2 | [] | 2026-10-21T00:00:00 | 2026-10-21T00:00:00 | 0 | 20 | training |
Hugging Face | [
"Hooking Face",
"hugging face"
] | [
"tech",
"ai_interactions"
] | the AI model and dataset hub | 2 | [] | 2026-10-21T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
My Weird Prompts | [
"MyWord Prompts"
] | [
"personal"
] | Daniel's AI podcast | 2 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Boxty | [
"E box d"
] | [
"personal"
] | null | 1 | [] | 2026-10-21T00:00:00 | 2026-10-21T00:00:00 | null | null | not-listed |
Carrot Cake AI | [
"Karrot Cake AI"
] | [
"personal"
] | a company/brand name Daniel uses, linked from his profiles | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Chatterbox | [
"Shatterbox"
] | [
"tech",
"ai_interactions"
] | Resemble AI's open-source TTS model | 1 | [
"shakshuka Shakshuka v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
eSpeak | [
"Espeak"
] | [
"linux"
] | the open-source speech synthesizer | 1 | [
"shakshuka Shakshuka v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
EXE | [
"AXE"
] | [
"tech"
] | the Windows executable file format, said as letters | 1 | [] | 2026-10-21T00:00:00 | 2026-10-21T00:00:00 | 0 | 20 | training |
Jupyter | [
"Jupiter"
] | [
"tech"
] | Open-source project behind Jupyter Notebook and JupyterLab; pronounced like the planet Jupiter | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Jupyter Notebook | [
"Jupiter notebooks"
] | [
"tech"
] | Interactive Python notebook environment (.ipynb files) for data science and ML | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Karpathy | [
"Karpathi"
] | [
"name",
"tech"
] | Andrej Karpathy, AI researcher and educator (ex-OpenAI, ex-Tesla AI) | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Kdenlive | [
"Caden Live"
] | [
"linux"
] | the KDE video editor | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Meorav | [
"Maorab"
] | [
"personal",
"hebrew"
] | Daniel's TTS voice-clone model; Hebrew for 'mixed' | 1 | [
"sabich Sabich v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
ONNX | [
"ONIX"
] | [
"tech",
"ai_interactions"
] | the Open Neural Network Exchange model format | 1 | [] | 2026-10-21T00:00:00 | 2026-10-21T00:00:00 | 0 | 20 | training |
prosody | [
"porosity"
] | [
"other"
] | the rhythm, stress and intonation of speech, as in TTS | 1 | [
"shakshuka Shakshuka v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Qwen | [
"Qwen1"
] | [
"tech",
"ai_interactions"
] | Alibaba's Qwen model family | 1 | [
"shakshuka Shakshuka v2.0"
] | 2026-10-22T00:00:00 | 2026-10-22T00:00:00 | 0 | 20 | training |
Sabich | [
"Tzabich"
] | [
"personal",
"hebrew"
] | Daniel's main ASR fine-tune, named after the Israeli sandwich | 1 | [] | 2026-10-21T00:00:00 | 2026-10-21T00:00:00 | 0 | 20 | training |
Hebrish ASR Error List
The words the Hebrish speech-to-text fine-tunes still get wrong in real dictation, what they were heard as, and how far targeted retraining on each one has got.
The errors come from one speaker's own dictation with the Hebrish desktop app (English with Hebrew code-switching, Jerusalem). Each word here is fed back into training as a priority word: the recording app serves sentences with it first until there are enough takes of it. No transcript text, audio or other content of the dictations is published, only the word, the wrong forms and the model.
- 20 words, 39 occurrences. Last review: 2026-10-22.
models/model_line: the Hebrish model that produced the error (Sabich, Shakshuka, …, with version);model_revis the exact weights' revision, recorded from app 0.7.3. Errors from before the app recorded the model have none.source: reportedmeans the speaker reported the error directly; it is not in the reviewed history.takes/target: recordings of the word made for retraining so far, against the number wanted.status:training(under target),trained(target reached; a later error means it still fails).
| Word | Heard as | Occurrences | Takes / target |
|---|---|---|---|
| ROCm | ROCM, Rockham, Rockin, RCM | 13 | 0 / 20 |
| Hebrish | HeBrish, Hebris, keybrush | 3 | 0 / 20 |
| Nemotron | Nematron, Nementron | 3 | 0 / 20 |
| F-Droid | FDroid | 2 | 0 / 20 |
| Hugging Face | Hooking Face, hugging face | 2 | 0 / 20 |
| My Weird Prompts | MyWord Prompts | 2 | 0 / 20 |
| Boxty | E box d | 1 | |
| Carrot Cake AI | Karrot Cake AI | 1 | 0 / 20 |
| Chatterbox | Shatterbox | 1 | 0 / 20 |
| eSpeak | Espeak | 1 | 0 / 20 |
| EXE | AXE | 1 | 0 / 20 |
| Jupyter | Jupiter | 1 | 0 / 20 |
| Jupyter Notebook | Jupiter notebooks | 1 | 0 / 20 |
| Karpathy | Karpathi | 1 | 0 / 20 |
| Kdenlive | Caden Live | 1 | 0 / 20 |
| Meorav | Maorab | 1 | 0 / 20 |
| ONNX | ONIX | 1 | 0 / 20 |
| prosody | porosity | 1 | 0 / 20 |
| Qwen | Qwen1 | 1 | 0 / 20 |
| Sabich | Tzabich | 1 | 0 / 20 |
Release tests
release_tests: each Hebrish model release scored on held-out read recordings of these words (test
split of the speaker's read-aloud corpus, never trained on), with the shipped int8 ONNX files and the
Hebrish desktop engine on CPU. exact: clips whose transcript has the word exactly as the reference
writes it; ci: ignoring case; heard_as_hits: a known wrong form still appears. Small counts while
the test recordings accumulate (1 in 5 of each priority word's sentences is recorded for test).
Maintained by Daniel Rosehill (Hebrish). Built by scripts/publish_error_list.py in the
Voice-Training-Recorder repo.
- Downloads last month
- -