Specify and evaluate speech data before you ask for a price
Use these ungated templates to compare vendors on the same requirements, inspect the documentation fields worth asking for, and keep held-out evaluation separate from training delivery.
Three practical starting points
These are scoping aids. The specimen is clearly marked and does not imply that every field is included in every delivery.
Speech dataset specification
Define languages, speakers, hours, capture conditions, annotations, rights, evaluation splits, and acceptance tests. Copy or download the generated brief.
Speech data card and provenance structure
Inspect a sanitized example covering provenance, consent references, capture, coverage, annotations, splits, QA, rights, and a manifest row.
Voice-agent evaluation brief
Scope conditions, held-out rules, reference data, slices, metrics, acceptance criteria, rights, and release decisions before collecting test audio.
Request samples for the buying job, not just the language
These offers are custom-scoped. Tell us the language and channel, then an admin reviews the request and sends a relevant sample manually.
Call-center and telephony
Consented simulated calls, phone-channel delivery, reference transcripts, and agreed scenarios.
Voice-agent evaluation
Representative conversational or telephony audio selected around your accent, noise, and failure conditions.
Full-duplex conversation
Separate-channel conversation examples with overlap, interruptions, backchannels, and turn timing.