metadata
license: cc-by-4.0
language:
- en
- he
pretty_name: Take Labels (Hebrish recordings)
tags:
- speech
- asr
- annotation
- noise
- pronunciation
configs:
- config_name: default
data_files: labels.jsonl
Take Labels
Hand labels for takes in danielrosehill/Read-Aloud-Voice-Sentences
and its subsets, set after recording by listening back. Join on id. 1 takes labelled (export of 2026-10-27).
A label overrides the value of the same thing recorded at capture time (for example a take tagged
quiet-indoor that was recorded on a street). null = not judged; an empty list = judged, none present.
| Field | Values |
|---|---|
environment |
quiet-indoor, noisy-indoor, outdoor, in-car |
noise_sources |
traffic, honking, light-rail, bus, motorbike, siren, construction, wind, pa-announcement, shouting, crying-baby, dog, appliance, tv, footsteps, dishes, other |
background_speech_languages, background_speech_level |
he, en, yi, ar, ru, am, fr, other, unknown; low / medium / high |
music_level, music_vocals, possible_copyrighted_music |
background music |
hebrew_pronunciation |
israeli, ashkenazi, anglicized, mispronounced, mixed: how the Hebrew words were said. Variation is deliberate: the speaker records Israeli and Ashkenazi forms, common Anglo-Israeli ways of saying a word and sometimes common mispronunciations, all transcribed with the same spelling |
microphone |
internal, bluetooth, wired-headset, external (the corpus column says internal on every row) |
phone_far |
phone held far from the mouth |
vocal_events |
cough, laugh, breath, throat-clear, lip-smack, sigh, false-start, hesitation |
take_quality, said_text |
ok, misread, cut-off, unusable; for a misread take, what was actually said |
tier |
golden (by hand); silver (classifier, not built yet) |
Part of the Hebrish project.