← Call Datasets

Fiund / Resource guide

A delivery specification for call audio datasets

Describe audio files, transcripts, labels, manifests, and exceptions as one connected package.

Specify audio artifacts

List acceptable formats, sampling rates, channel arrangements, and source-preservation requirements. Distinguish original recordings from processed versions. Define the call or interaction identifier and filename rules, but avoid placing unnecessary personal details in paths. Agree on whether linked call legs remain separate or are delivered with a documented relationship map.

Connect transcripts and labels

Choose machine-readable schemas for text, timestamps, speaker segments, roles, and any task labels. Include units, uncertainty markers, and version identifiers. Every annotation should resolve to the correct recording or channel without depending on an undocumented folder convention or a manually maintained spreadsheet description.

Include a useful manifest

Request file sizes, durations, checksums when agreed, source subset, and relevant coverage fields. Summarize missing artifacts, corrupt files, unresolved labels, and excluded material separately. Distinguish raw source totals from reviewed project material so a delivery inventory does not imply that all listed recordings meet every requirement.

Reconcile a pilot first

Open representative files, trace labels to audio, and compare duration totals across the package. Confirm channel maps and timing behavior before scaling the workflow. Send Fiund the delivery schema and acceptance questions early so prospective source and preparation plans can be assessed against an explicit, project-specific definition of completeness. Test the package with a consumer unfamiliar with the preparation workflow. Ask them to locate one call, identify its channels, open its transcript, and inspect a speaker segment using only the manifest and documentation. Their unresolved questions provide a practical delivery checklist before the same conventions are applied across more material.

Review checklist

  • Choose original and derived audio requirements
  • Define linked annotation schemas
  • Include manifests and exception summaries
  • Reconcile representative artifacts before expansion

Your project, in focus

Tell us which conversations your voice model needs.

Speak directly with Jaeden Schafer at Fiund about the source material, specifications, and review your project needs.

Source fit, permissions, samples, pricing, and delivery are reviewed for the individual project.

Start with these details

  1. Languages, dialects, and regions
  2. Call scenarios and real-versus-staged requirements
  3. Hours, call lengths, and channel layout
  4. Transcript and redaction specifications
  5. Intended model use and licensing terms
Build your project brief ↗