Start with the speech task
Speech recognition, speaker diarization, and conversational modeling emphasize different properties. List the languages, interaction styles, recording conditions, and text annotations your project needs. Explain whether spontaneous dialogue is essential or whether scripted speech can also fit.
Distinguish show diversity from speaker diversity
Many episodes can share the same host and production setup. If the aim is broader coverage, ask for distinct speakers, recording sessions, shows, and environments. Describe how recurring guests and repeat appearances should be counted.
Choose the recording level
A finished episode may include edits, advertisements, music, and other inserted material. Original session tracks can offer different properties but may have pauses or production chatter. Specify the exact delivery level and the transformations that are acceptable.
Review the rights for this use
Ask what documentation covers the show, participants, and any included third-party material. Establish exclusions before samples are distributed. Discuss the proposed training use, retention, attribution, and sublicensing requirements as part of the commercial review.
Review checklist
- Languages, topics, and conversation styles
- Hours and preferred speaker diversity
- Mixed audio or independent tracks
- Transcript, diarization, and edit exclusions
- Training use and participant-permission requirements