Dataset Viewer
Auto-converted to Parquet Duplicate
audio
audio
id
string
text
string
duration_s
float64
categories
list
target_terms
list
prompt_id
string
prompt_source
string
mode
string
edited
bool
device
string
environment
string
sample_rate
int64
recorded_at
string
microphone
string
prompt_origin
string
hebrew_pronunciation
string
variant_groups
list
variant_spelling
string
speaker_id
string
conditions
list
imported_from
string
original_text
string
text_before_normalization
string
text_he
string
noise_level
string
background_speech_languages
list
notes
string
model_hypothesis
string
asr_model
string
speaker
string
speaker_gender
string
noise_flags
list
clipped
bool
task
string
spoken_text
string
aside_text
string
aside_position
string
verbatim_text
string
prompt_categories
list
source_prompt_id
string
prompt_kind
string
prompt_note
string
term_language
string
paragraph_text
string
contributor
string
bit_depth
int64
audio_format
string
native_rate
int64
capture_mode
string
tts_eligible
bool
tts_split
string
transcript_status
string
prompt_set
string
topic
string
style
string
style_instruction
string
written_form_uncertain
bool
session_id
string
room_tone_id
string
lead_trim_s
float64
peak_dbfs
float64
floor_dbfs
float64
speech_dbfs
float64
snr_db
float64
clipped_samples
int64
edge_head_dbfs
float64
edge_tail_dbfs
float64
browser_processing
bool
room
string
priority_word
string
location
string
location_name
string
location_note
string
latitude
float64
longitude
float64
location_accuracy_m
float64
text_full
string
fallback_ok
bool
sentence_id
string
pronunciation
string
extra
string
20261120T060811Z_19n6wx
Before I forget, pick up the dry cleaning, the stamps, and a birthday card for Auntie Jo on the way home.
5.338
[ "run-on" ]
[]
run-on_70ec5d4880
generated:claude-self-written
bank
false
OnePlus Nord 3 5G
quiet-indoor
48,000
2026-11-20T06:08:11.295Z
internal
generated
null
[]
null
8992
[]
null
null
null
Before I forget, pick up the dry cleaning, the stamps, and a birthday card for Auntie Jo on the way home.
null
[]
null
null
null
Daniel Rosehill
male
[]
false
run-on
Before I forget, pick up the dry cleaning, the stamps, and a birthday card for Auntie Jo on the way home.
null
null
null
null
null
control
null
null
null
Daniel Rosehill
16
flac
48,000
stt
true
null
script
null
null
null
null
null
20261120T055558Z_3xm44l
null
0
-3.4
-45.1
-16.4
28.8
0
-30.7
-39.3
false
null
null
null
null
null
null
null
null
null
null
p_4708a9fc7a
[]
{"mic_link":null,"bt_codec":null,"mic_name":"CPH2493","mic_usb_vendor_id":null,"mic_usb_product_id":null,"mic_type":"builtin_mic","mic_chain":null,"activity":"seated","background_speech_kinds":[],"pronunciation_variants":[],"ui_note":null,"source_url":null,"check_verdict":null,"check_errors":null,"check_error_types":nu...

Run-On Sentences

Type: Audio Role: Split-out Subset of: Read-Aloud-Voice-Sentences

Audio dataset, split-out. Rows copied from the master, danielrosehill/Read-Aloud-Voice-Sentences, so one task can be browsed alone. Training reads the master, not this copy. Every audio dataset as one tree, with hours: the audio index. How every dataset fits together: the dataset index.

One speaker reading run-on dictation in one breath: complete sentences joined by commas, "so" or "and". spoken_text is the comma-joined card as read; text writes the separate sentences, with the same words in the same order (only punctuation and first-letter case differ). Controls (prompt_kind control) are clauses that must stay in one sentence and keep their commas. Experimental. Prompts written by Claude Sonnet (2026-11-06) from Daniel's most frequent hand edit, comma to full stop.

Website: hebrish.app: get the Hebrish keyboard for Android and the desktop apps, see the results, and find every model, dataset and demo in one place.

Two tabs. sentences lists every ground-truth sentence with a takes count, so takes = 0 shows what has not been recorded. takes has one row per recording with its audio, joined on sentence_id; a sentence can have several takes. load_dataset("Hebrish/Run-On-Sentences", "sentences") reads the text; the default config reads the takes.

Takes come from danielrosehill/Read-Aloud-Voice-Sentences, the full dataset, with the same ids and train/test split, by scripts/sync_subsets.py in the project repository. Do not edit this dataset directly. New takes append and only the sentence counts are rewritten.

Speaker: Daniel Rosehill, the dataset's author, who was born in Ireland and speaks English with an Irish accent, and who lives in Jerusalem. Every recording is his voice. Deliberate variation: he sometimes changes his speaking speed on purpose (fast and rushed, or slow) and sometimes says a word the way people commonly mispronounce it, to make the models robust to real speech. The transcript is always the sentence as written, so a take whose speed or pronunciation sounds off against its card is usually intentional, not a recording error.

All audio was recorded on a OnePlus Nord 3 5G phone using its internal microphone, with browser echo cancellation, noise suppression and auto gain turned off.

Layout

sentences/train.parquet  one row per sentence: `id`, `text`, `takes`, `prompt_ids`, `served` and the sentence columns
data/train-*.parquet   one row per take: `audio` (the FLAC or WAV file, embedded) and every column
meta/train.jsonl       index: the same rows without audio, plus `shard`, `audio_path` and the audio's byte range
meta/schema.json       column types
data/test-*.parquet, meta/test.jsonl   held-out set, never trained on

Audio is 48 kHz mono 24-bit FLAC for rows from 2026-10-22 and 16 kHz mono 16-bit WAV before.

The test split is term-disjoint: a held-out term never appears in train. load_dataset("Hebrish/Run-On-Sentences") works directly.

Columns

A take's sentence_id is its row's id in sentences. A sentence with served: false is no longer offered by the app; it is kept because a take used it. The columns are the same as in the main dataset: text (ground truth, punctuated and cased like Parakeet v2 output), text_he (the same transcript with its Hebrew terms in Hebrew script, for Hebrew-capable models), categories, target_terms, prompt_source, prompt_origin (manual / dataset / generated), edited, duration_s, environment, conditions, device, speaker, speaker_gender, contributor, hebrew_pronunciation, variant_groups, variant_spelling and recorded_at. The main dataset's card describes them.

Licence

CC-BY-4.0, with sentence-text attribution per prompt_source as described in the main dataset's licence section.

Downloads last month
77