A build-to-spec dataset of human preferences for candidate voice responses, with rankings, spoken rationales, and evaluation metadata defined for each program.
Datoric’s RLHF-Voice Dataset is a build-to-spec collection of human preferences for voice-model responses. A comparison may include a prompt, two or more candidate responses, a human ranking or pairwise preference, and—when specified—spoken rationales and transcripts. Programs define their target models, languages, reviewer mix, rubric, and acceptance criteria during scoping. The described data package can include WAV or PCM rationale audio, CSV preference records, JSON metadata, a data dictionary, and QA summaries. Datoric says commercial licensing and final delivery scope are set by agreement.
Be the first Cosign
Add it to your Products I use list. You can also share why you use it.
