Datasets:
Download README.md from danielrosehill/Scripted-Voice-Notes: direct link, hf CLI and curl.
- Browser
- Download file 5.78 kB
-
https://huggingface.co/datasets/danielrosehill/Scripted-Voice-Notes/resolve/main/README.md
- Command line
-
hf download hf://datasets/danielrosehill/Scripted-Voice-Notes/README.md
-
curl -L -o README.md https://huggingface.co/datasets/danielrosehill/Scripted-Voice-Notes/resolve/main/README.md
license: cc-by-4.0
language:
- en
task_categories:
- automatic-speech-recognition
pretty_name: Scripted Voice Notes
tags:
- speech
- asr
- fine-tuning
- single-speaker
- voice-notes
- dictation
- intent
- english
size_categories:
- 1K<n<10K
configs:
- config_name: takes
default: true
data_files:
- split: train
path: data/train-*.parquet
- config_name: sentences
data_files:
- split: train
path: sentences/train.parquet
Scripted Voice Notes
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 recording voice notes from scripts: the text is synthetic, the voice is real. Notes of the kinds a voice note -> action assistant acts on (email drafts, blog and social post drafts, calendar entries, new contacts, to-dos, shopping lists, meeting notes, AI prompts, development specs, notes to self, ideas, some with two actions), and finished everyday messages dictated in one take (emails at four formality levels, calendar instructions, shopping and task lists, WhatsApp, Slack and Teams messages, journal entries). categories names the source task: voice-actions rows carry the route labels of danielrosehill/Voice-Note-Task-Classifier (prompt_kind the primary label, prompt_note the full label list, style and scenario); voice-notes rows have the message type in prompt_kind. Scripts written by DeepSeek V4.1 Flash, not reviewed (ASR-Training scripts/generate_voice_actions.py and generate_voice_notes.py, 2026-11-15/16). The takes are copies: the same takes are also in Hebrish/Voice-To-Action and Hebrish/Voice-Notes.
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("danielrosehill/Scripted-Voice-Notes", "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("danielrosehill/Scripted-Voice-Notes") 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.