Long conversations
Natural, extended, full-duplex conversation data between two speakers.

From everyday conversations to specialist domains, built around the way people actually speak.
Natural, extended, full-duplex conversation data between two speakers.

Environments built using real human speech to improve instruction following and conversational handling.
Speech across diverse languages and accents, spanning more than 100 countries and locales.
Natural conversations between three or more speakers.
Conversations or script reading in finance, medicine, education, and other specialist fields.
Professional voice actor recordings with a custom commercial license.
Have interesting data you can license? Get in touch with us!
Explore recording formats you can license for model training and evaluation. Browse the inventory for language coverage, recording specifications, and audio samples.
Long-form, unscripted dialogue with synchronized speaker tracks that preserve overlap, interruptions, and backchannels.
Two people act out a service call, from the opening question to a resolution. Separate speaker tracks preserve turn-taking and overlap.
Read speech across accents and specialist domains, including medicine and finance, for speech recognition and domain vocabulary.
Scripts with spoken email addresses, URLs, dates, and commands for building and testing models that must transcribe exact values.
Tell us what your model needs to hear or do. We can record to your brief, label audio you already have, or build new evaluations for your voice agents.
Speech recordings collected to your language, speaker, and scenario requirements.
Amateur voice actors who can read and converse in natural, non-performative, emphatic voices. Great for text-to-speech model and speech-to-speech model training.
Transcripts reviewed and corrected by human annotators.
Annotations built around your labels, guidelines, and quality requirements.
Compare model outputs with human judgments, run ELO scoring, or focus on specific locales.
Scenarios and tasks for voice agents with custom designed system prompt, constraints, rubrics, real recorded conversations, and turn by turn scoring.