Skip to content

4. Custodians: Rights and Responsibilities

[!NOTE] In a nutshell, custodians can use audio recordings and human annotations for model development, but not to answer substantive scientific research questions unrelated to model performance (e.g., cross-linguistic variation in child-directed speech, language acquisition trajectories). Such uses require corpus collector notification and offer of co-authorship or data withdrawal (see Section 5). The custodians also make sure to comply with best practices regarding data security.

Rights

  • Access raw audio recordings and human annotations from member datasets for the development, training, evaluation, and benchmarking of automated speech and language processing models (e.g., child vocalization detection, speaker diarization, language identification).
  • Publish results of model development and benchmarking research in academic venues (journals, conference proceedings) with appropriate citation of all data sources, without requiring corpus collectors as co-authors, provided the research falls within the defined scope.
  • Apply automated models to member datasets and retain derived annotations for research purposes.

Responsibilities

  • Ascertain that all supervisees coming in contact with data are highly aware of the sensitive nature of the recording data and agree to uphold high levels of data security.
  • Follow procedures for declaring supervisees as defined by each scientific archive.
  • Open-source models developed using consortium datasets when feasible and consistent with applicable intellectual property policies; in particular, the license must exclude commercial use for compatibility with rules of the scientific archives.
  • Notify corpus collectors and offer co-authorship or data withdrawal when proposed research extends beyond model development and benchmarking.
  • Ensure proper citation and acknowledgement of all data sources in every publication.
  • Comply with the legal and ethical access conditions of the scientific archives through which data access is granted.
  • Decline all third-party requests for access to corpus collector data and direct such requests to the relevant corpus collector and/or archive (see Section 6).