01
There are two Open Yap 1K access paths
The public Hugging Face repository contains an 8.9-hour sample under CC BY 4.0. The full 1,000-hour corpus is available separately by request under the Open Yap 1K Data Use Agreement.
Do not treat the Hugging Face license badge as the license for all 1,000 hours. The sample and full corpus have different access and licensing terms.
02
1. Download the Hugging Face sample
The sample contains 16 conversations, 8.9 hours and eight speakers. The repository includes both synchronized speaker tracks, transcripts and metadata and is listed under CC BY 4.0.
The publisher documents downloading the WebDataset shards with the Hugging Face CLI using the dataset repository TheAgenticDataCompany/open-yap-1k. Recheck the live repository before automating around exact file names because repository structure can change.
03
2. Request the full 1,000-hour corpus
For the complete corpus, submit an access request explaining who you are, what you are building and how the audio will be used. The publisher reviews requests and says approved recipients receive portal access.
Delivery is currently described as either a direct download or transfer into the recipient's own S3 bucket. The publisher says replies are usually sent within a few hours, which should be treated as a vendor-stated expectation rather than a guaranteed SLA.
04
3. Keep the sample and full-corpus licenses separate
The Hugging Face sample is listed under CC BY 4.0. The full corpus uses the Open Yap 1K Data Use Agreement and is currently offered at no charge for commercial and research use.
The full agreement permits training, fine-tuning, evaluation, benchmarking, deployment and commercialization of models and outputs, subject to its restrictions.
05
4. Understand the restrictions
- do not redistribute, reshare, sublicense or resell the full dataset
- do not attempt to identify a speaker or link a recording to external records
- do not create voice clones or generative reproductions identifiable as a corpus speaker
- protect the dataset at least as strongly as your own confidential material
06
5. Handle team and contractor access carefully
The agreement permits internal copies and access for staff and contractors operating under the same terms. A practical deployment should keep the raw corpus in controlled storage and grant access only to people who need it.
Avoid mirroring the raw corpus into public buckets, model repositories or broadly shared drives. Keep the applicable agreement version alongside the data so the licensing boundary survives internal handoffs.
07
6. Account for data-quality boundaries
The transcripts are machine-generated with Deepgram Nova-3 and are not human-verified. The public sample is hand-picked, so its distributions should not be assumed to represent the full corpus.
The publisher also notes that some sample tracks are stored in 48 kHz containers even when Bluetooth recording hardware does not preserve usable energy across the full theoretical bandwidth. Inspect effective-bandwidth metadata before making signal-quality assumptions.
08
Bottom line
Use the Hugging Face sample for quick inspection and prototyping. Request the full corpus when you need the 1,000-hour release and are prepared to operate under the separate Data Use Agreement.
For the broader dataset structure, conversational statistics, use cases and limitations, return to the main Open Yap 1K overview.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.