Request access to Instinct speech data
Requests are reviewed manually. Approval is not automatic and may be revoked if the dataset terms are violated.
By submitting this request, you agree to use the dataset only for the stated purpose, follow its license and privacy terms, and not redistribute the data.
Log in or Sign Up to review the conditions and access this dataset content.
Instinct Speech Data Catalog
This is the machine-readable entry point for the instinct-org speech data estate. It contains
metadata only; source audio remains in its existing manually gated repositories.
Inventory
- Data products: 113
- Public/manual-gated data products: 109
- Private/manual-gated exceptions: 4
- Hub-reported storage: 7.81 TiB
Purpose
| Purpose | Datasets |
|---|---|
mixed |
1 |
stt |
56 |
tts |
55 |
unknown |
1 |
Lifecycle state
| State | Datasets |
|---|---|
aligned |
16 |
model-encoded |
18 |
packaged-pair-candidate |
12 |
pair-plan |
13 |
restored |
15 |
segmented |
10 |
source-unchunked |
11 |
tts-candidate |
15 |
unclassified |
3 |
Files
catalog.yaml: canonical human-readable registry.catalog.parquet: analytics-friendly registry.language_audit/language_samples.jsonl: gated language evidence with up to 10 transcript samples per dataset, the selected text field, observed language values, and script profile.language_audit/summary.yaml: language-audit coverage and status counts.schemas/: catalog, STT, and TTS schema contracts.policies/: explicit visibility, gating, and license policies.migrations/legacy-map.yaml: current names and proposed canonical names. Proposals are not renames.
Access
All public datasets require manual approval. Four explicitly registered SIDON datasets remain private and manually gated. The catalog does not grant access to any underlying dataset.
Language evidence
Declared language metadata is recorded in the catalog. Because transcript excerpts come from
gated datasets, the samples are stored in the gated language_audit/language_samples.jsonl file
rather than copied into publicly visible README prose. Each dataset has up to 10 unique samples;
missing text fields and access failures are recorded explicitly.
Lifecycle model
Purpose (stt, tts, or mixed) is separate from lifecycle state (source-unchunked,
segmented, restored, aligned, model-encoded, tts-candidate,
packaged-pair-candidate, or pair-plan). Do not infer lineage from repository names; use
catalog revisions and parents.
- Downloads last month
- -