The Agentic Data CompanyOpen Yap 1K
Request access
Version v1

Open Yap 1K Data Use Agreement

In plain English

Permitted
  • Commercial use, including in products you sell
  • Research use, published or internal
  • Training, fine-tuning and evaluating models, and deploying what you train
  • Internal copies, and access for staff and contractors under the same terms
  • Retaining the delivered dataset, and anything derived from it, after a speaker withdraws
Not permitted
  • Redistributing, resharing, sublicensing or reselling the dataset or any part of it
  • Attempting to identify a speaker, or link a recording to any external record
  • Creating voice clones, replicas or generative reproductions identifiable as a speaker in the corpus
  • Holding the dataset under weaker controls than your own confidential material
  • Retaining any copy of the dataset after a breach of these terms, or after a written request from The Agentic Data Company

A summary, not the agreement. The full text below governs.

OPEN YAP 1K DATA USE AGREEMENT
Version 1

This agreement governs access to Open Yap 1K, a corpus of English two-speaker
conversational speech published by The Agentic Data Company. It is offered on
the same terms to every recipient, at no charge, for both commercial and
research use.

1. Parties and acceptance
   This agreement is between The Agentic Data Company, a Delaware corporation
   with its registered office at 131 Continental Dr, Suite 305, Newark, New
   Castle County, Delaware 19713, United States ("the Company") and the
   organisation on whose behalf access is requested ("the Recipient"). The
   person completing the access request confirms that they are authorised to
   accept these terms for the Recipient and that the information in the request
   is accurate. Where no organisation is named, the person accepting is the
   Recipient in a personal capacity.

   Acceptance is given electronically, by marking the acceptance box on the
   access request form, and the parties intend it to have the same effect as a
   signature. Clause 24 sets out what the Company records as evidence of it.

2. Definitions
   "Dataset" means Open Yap 1K as delivered to the Recipient: the audio
   recordings, transcripts, speaker and conversation metadata, per-conversation
   metrics, manifest and accompanying documentation, in whole or in part, in
   any format.

   "Derived Materials" means anything the Recipient creates from the Dataset,
   including extracts, subsets, transformed or re-encoded audio, annotations,
   alignments, features, embeddings, statistics and Models.

   "Model" means a machine learning model, and the weights, adapters,
   embeddings or other parameters of one, trained or tuned in whole or in part
   on the Dataset.

   "Output" means content a Model generates.

   "Speaker" means an individual whose voice appears in the Dataset.

   "Personnel" means the Recipient's employees, and the contractors and
   individual consultants working on the Recipient's behalf.

   "Affiliate" means an entity that controls, is controlled by, or is under
   common control with a party.

3. Grant
   The Company grants the Recipient a worldwide, non-exclusive,
   non-transferable, non-sublicensable, royalty-free licence to use, copy and
   modify the Dataset for commercial or research purposes, subject to this
   agreement. No fee is payable for the Dataset.

   The licence extends to the Recipient's Affiliates. The Recipient remains
   responsible for each Affiliate's compliance with this agreement, and a
   breach by an Affiliate is a breach by the Recipient. An entity that ceases
   to be an Affiliate stops being licensed, and clause 13 applies to it.

   The Company is under no obligation to deliver further data, to correct the
   Dataset, to support it, or to keep it available.

4. Permitted use
   Within the limits of clauses 5 to 9, the Recipient may:

   (a) use the Dataset internally for any commercial or research purpose,
       including developing products it sells;
   (b) train, fine-tune, adapt, evaluate, benchmark and test Models on the
       Dataset, including generative speech, speech-to-speech, recognition,
       diarization and audio understanding models;
   (c) deploy, license and commercialise those Models and their Outputs;
   (d) create and use Derived Materials for those purposes.

   The Company claims no ownership of a Model the Recipient trains or of its
   Outputs. A Model or an Output is not restricted by clause 8 merely because
   the Dataset was among the data a Model was trained on.

5. No redistribution
   The Recipient may not publish, share, sublicense, sell, lend or otherwise
   make the Dataset, or any part of it, available to any third party, and may
   not deposit it in a public repository, a shared research archive, a model
   hub, or any other third-party repository.

   This restriction applies to Derived Materials to the extent they contain,
   reproduce, or allow the reconstruction of Dataset audio, transcripts or
   speaker metadata. It does not restrict publication of Models, Outputs,
   aggregate statistics, or research results that do not.

   Short transcript excerpts, of no more than a few sentences and drawn from no
   more than a small number of conversations, may be quoted in a publication or
   presentation where necessary to illustrate a finding. Publishing Dataset
   audio, in any length or form, requires the Company's prior written consent.

6. Internal copies, personnel and contractors
   The Recipient may make the copies of the Dataset its permitted use requires,
   including backups and copies held on cloud or compute infrastructure the
   Recipient controls or has engaged under a written contract.

   The Recipient may give access to Personnel who need it for a permitted use,
   and to no one else. It must ensure each of them is bound by confidentiality
   and use obligations at least as protective as this agreement, remains
   responsible for their acts and omissions as for its own, and withdraws
   access when the need ends.

   The Recipient must keep a record of where copies of the Dataset are held,
   sufficient to carry out clause 13.

7. No re-identification
   The Recipient may not attempt to identify any Speaker, and may not attempt
   to link any recording, transcript or item of metadata to a name, an account,
   a contact detail, an external voice sample, or any other external
   identifier, record or dataset. It may not attempt to reverse the
   pseudonymous speaker and conversation identifiers the Dataset uses.

   Using the Dataset's own speaker identifiers to train or evaluate speaker
   verification, speaker diarization or similar models is permitted. It is not
   re-identification, because it does not seek the identity of the person
   behind the identifier.

   If the Recipient learns a Speaker's identity, incidentally or from
   identifying information left in a recording or transcript, it must not act
   on or record that knowledge and must notify the Company promptly under
   clause 21.

8. No voice cloning
   The Recipient may not use the Dataset to create, and may not configure a
   Model to produce, a synthetic voice, voice replica, or generative
   reproduction of a person's voice or likeness that a reasonable person would
   identify as matching an individual in the Dataset.

   In particular, the Recipient may not:

   (a) train, tune or condition a Model on the recordings of a single Speaker,
       or of a selected group of Speakers, in order to reproduce that voice;
   (b) use a Dataset recording as a reference, prompt or target for voice
       conversion or zero-shot voice cloning;
   (c) publish, sell or otherwise make available a voice, voice preset or
       speaker embedding derived from an individual Speaker.

   This clause is about targeting an individual. Training a general Model on
   the Dataset alongside other data, and deploying it, is permitted by clause
   4, and an incidental resemblance in an Output that the Recipient did not
   seek is not a breach of this clause. The Recipient must not deliberately
   elicit such an Output, and on becoming aware that one of its Models
   reproduces an identifiable Dataset voice, must take reasonable steps to stop
   that recurring and notify the Company.

9. Security
   The Recipient must hold the Dataset under administrative, technical and
   physical controls no weaker than those it applies to its own confidential
   material, and in any event under controls reasonable for personal data,
   including access control, encryption in transit and encryption at rest.

   The Recipient must notify the Company without undue delay, and no later than
   72 hours, after becoming aware of any loss of the Dataset or any access to
   it that this agreement does not permit.

10. Confidentiality
   The Dataset is the Company's confidential information, and the Recipient may
   not disclose it except as clause 6 allows. This does not apply to
   information that is or becomes public other than through the Recipient, that
   the Recipient already held free of any duty of confidence, or that it
   develops independently.

   The Recipient may disclose the Dataset where required by law or court order,
   and must, where lawful, give the Company prior notice so it can seek
   protection. The existence of this agreement and the fact of the Recipient's
   access are not confidential.

11. Personal data
   The Dataset contains personal data: recorded voices, and demographic
   information the Speakers reported about themselves. It is pseudonymised, not
   anonymous. Names, contact details and account identifiers are not included.
   Every Speaker gave explicit consent before recording and was paid for their
   time.

   The Recipient must handle the Dataset in accordance with the data protection
   law that applies to it, and must not use it in a way inconsistent with that
   consent or with clauses 7 and 8.

   The parties act as independent controllers in respect of the personal data
   in the Dataset. Neither party processes that personal data on behalf of the
   other, and neither is a processor or joint controller of the other. Each
   determines its own purposes and means for the personal data it holds, and
   each is separately responsible for its own compliance with the data
   protection law that applies to it, including for responding to the requests
   of data subjects it receives.

   The Company remains responsible for the lawfulness of the collection and of
   the disclosure to the Recipient. The Recipient is responsible for its own
   processing from the point of delivery onward.

   Where a delivery to the Recipient requires a transfer instrument under the
   data protection law applying to the Company, the Company may require the
   Recipient to enter into the European Commission's standard contractual
   clauses for controller-to-controller transfers, or another instrument the
   Company reasonably specifies, as a condition of delivery. Those clauses
   prevail over this agreement to the extent of any conflict.

12. Retention and speaker withdrawal
   The licence granted by clause 3 over a delivery already made is perpetual.

   A Speaker may withdraw from the Company's platform at any time. Withdrawal
   ends further collection from that Speaker and further distribution of their
   recordings by the Company. It does not require the Recipient to delete, stop
   using, or retrain anything already delivered, or any Derived Materials.

   The Company may exclude a Speaker's recordings from later deliveries or
   later versions of the Dataset without notice.

13. Deletion and certification
   The Recipient must permanently delete all copies of the Dataset, and all
   Derived Materials that contain or reproduce Dataset audio, transcripts or
   speaker metadata:

   (a) if it breaches this agreement; or
   (b) on written request from the Company, within 30 days of the request.

   Deletion covers copies held by Personnel and copies on infrastructure
   engaged under clause 6. The Recipient may retain copies it is required by
   law to keep, and copies in routine backups that cannot practicably be
   isolated, for as long as those constraints last; this agreement continues to
   apply to them.

   Models trained before the deletion event, and Derived Materials that do not
   contain or reproduce Dataset content, may be retained and used, except where
   the Model or material is itself the subject of a breach of clause 7 or 8, in
   which case it must be destroyed. Clause 8 continues to apply to a Model
   retained under this paragraph, whenever it was trained.

   Within 30 days of completing a deletion the Recipient must confirm in
   writing, signed by an officer, that it has done so, and describe anything
   retained under this clause.

   Separately, once in any twelve month period the Company may ask the
   Recipient to confirm in writing that it is complying with clauses 5 to 9.
   This agreement gives the Company no right of on-site audit and no right of
   access to the Recipient's systems.

14. Attribution
   Published work that makes use of the Dataset should cite it using the
   reference published alongside the Dataset, including the version identifier
   of the build used. The Company publishes that reference, and a BibTeX entry
   for it, on the Dataset page.

15. Compliance
   The Recipient must comply with the laws applying to its use of the Dataset,
   including data protection, privacy, export control and sanctions law.

   The Recipient may not use the Dataset to build or operate a system whose
   purpose is unlawful surveillance of individuals or unlawful discrimination
   against them.

16. No warranty
   The Dataset is provided "as is" and "as available". To the maximum extent
   permitted by law the Company disclaims all warranties, express or implied,
   including merchantability, fitness for a particular purpose, accuracy,
   completeness, quiet enjoyment and non-infringement.

   Transcripts are machine generated and are not human verified. Demographic
   fields are reported by Speakers about themselves and are not verified
   against any external record. Published quality figures describe the corpus
   in aggregate and are not a warranty about any individual file. The Dataset
   is screened for personally identifying information, but the Company does not
   warrant that none remains.

17. Limitation of liability
   Nothing in this agreement excludes or limits either party's liability for
   death or personal injury caused by its negligence, for fraud or fraudulent
   misrepresentation, for gross negligence or wilful misconduct, or for any
   other liability that cannot lawfully be excluded or limited. The exclusions
   and the cap in this clause are subject to this paragraph.

   To the maximum extent permitted by law, the Company is not liable for any
   indirect, incidental, special, consequential, exemplary or punitive damages,
   or for lost profits, lost revenue or loss of data, arising out of or
   relating to this agreement or the Dataset, whether in contract, tort or
   otherwise, and whether or not it was advised such loss was possible.

   The Company's total aggregate liability arising out of or relating to this
   agreement is limited to one hundred United States dollars (USD 100). The
   Dataset is supplied at no charge, and this figure reflects that.

   This clause does not limit the Recipient's liability for breach of clause 5,
   7, 8 or 10, or its obligations under clause 18.

18. Indemnity
   The Recipient will defend, indemnify and hold harmless the Company and its
   officers, employees and agents against any third-party claim, and against
   the liabilities, damages, fines, penalties and reasonable legal costs
   arising from it, to the extent the claim arises out of the Recipient's use
   of the Dataset in breach of this agreement, its breach of clause 7, 8, 9 or
   15, or a claim brought by a Speaker arising from either.

   The Company will notify the Recipient of the claim without undue delay, let
   the Recipient control its defence, and give reasonable cooperation at the
   Recipient's expense. The Recipient may not settle a claim in a way that
   admits fault by the Company or imposes an obligation on it without the
   Company's written consent.

   Public institutions. Where the Recipient is a public university, a public
   research institution, a government body or another public entity, and is
   prohibited by the constitution, a statute, or a binding public policy of
   its jurisdiction from giving the indemnity in this clause, that indemnity
   does not apply to it and the two paragraphs above do not bind it. In their
   place the Recipient covenants to use the Dataset only as this agreement
   permits and to comply with clauses 7, 8, 9 and 15. For such a Recipient the
   Company's remedies for breach are termination under clause 19 and
   injunctive or other equitable relief, together with any remedy available to
   the Company under the law that applies; and the Recipient's liability, if
   any, is that provided by the law that applies to it rather than an
   obligation to indemnify or defend the Company. A Recipient relies on this
   paragraph as the public institution it names in its access request.

19. Term, termination and survival
   This agreement starts when the Recipient accepts it and continues until
   terminated.

   Either party may terminate for convenience on 30 days written notice. The
   Company may terminate immediately on notice if the Recipient breaches clause
   5, 7, 8, 9, 10 or 15, or fails to remedy any other breach within 30 days of
   being asked to.

   Termination ends any right to further deliveries. It does not by itself
   require deletion: the perpetual licence in clause 12 survives termination
   for convenience, and deletion is required only in the circumstances set out
   in clause 13.

   Clauses 2 to 18 and 20 to 24 survive termination. The licence in clause 3
   survives only over deliveries already made, on the terms of clause 12.

20. Assignment
   The Recipient may not assign or transfer this agreement, or any right under
   it, without the Company's prior written consent, except to a successor to
   all or substantially all of its business or assets, on written notice to the
   Company. The Company may assign this agreement freely. Any other purported
   assignment is void, and the Dataset may not be transferred as part of an
   asset sale except under a permitted assignment.

21. Notices
   Notices to the Company go to The Agentic Data Company, 131 Continental Dr,
   Suite 305, Newark, New Castle County, Delaware 19713, United States, and by
   email to legal@theagenticdatacompany.com. Notices to the Recipient go to
   the email address given in the access request, or to a later address the
   Recipient gives the Company in writing. A notice sent by email is given
   when sent, unless the sender receives a delivery failure.

22. Governing law and venue
   This agreement is governed by the laws of the State of Delaware, United
   States, without regard to its conflict of laws rules. The state and federal
   courts sitting in Delaware have non-exclusive jurisdiction over any dispute
   arising out of or relating to this agreement, and each party submits to that
   jurisdiction. The United Nations Convention on Contracts for the
   International Sale of Goods does not apply.

   Nothing in this clause limits any right a Speaker has under the data
   protection law of their own jurisdiction, or any obligation the Company owes
   under it.

23. General
   This agreement, together with the access request given with it, is the
   entire agreement between the parties about the Dataset and replaces any
   earlier discussion or understanding about it. A purchase order or a
   recipient's own standard terms have no effect.

   The Company may issue a revised version of this agreement for later
   deliveries. A revision does not change the terms accepted for a delivery
   already made; the version accepted at the time of a delivery governs that
   delivery.

   If a provision is held unenforceable it is limited to the minimum extent
   necessary and the rest stays in force. A failure to enforce a right is not a
   waiver of it. The parties are independent contractors, and this agreement
   creates no rights for anyone other than them.

24. Acceptance record
   Acceptance is recorded when the Recipient submits the access request with
   the acceptance box marked. The Company records the version of this agreement
   accepted, a SHA-256 hash of the exact text served at that moment, the
   answers given in the request, and the time, network address and browser user
   agent of the submission.

   That record is how the Company evidences what was accepted once this text is
   revised. The Company will keep each accepted version retrievable at a stable
   address.