Multimodal datasets
Paired audio, image, video and text built for models that read more than one signal.
What we hold in multimodal
Off-the-shelf sets listed below. Anything not on the list can be collected to your spec, usually inside four weeks.
- Image and caption pairs
- Audio with aligned transcript and video
- Document image with structured output
- Screen recording with intent labels
- Sensor fusion for driving
- Speech with speaker and emotion tags
3 multimodal datasets available
3 of 3 datasets
- MultimodalMultimodal
Visual question answering
Image and question pairs with grounded answers and reasoning traces.
12 languages
Details - MultimodalMultimodal
Image and caption pairs
Descriptive captions written by native speakers, not machine translated.
15 languages
Details - MultimodalDocument AI
Document image to structured output
Document images paired with the structured JSON a model should produce.
10 languages
Details
Other dataset types
Speech datasets
Conversational, scripted, wake-word and call-centre audio with verified transcripts.
Learn more
Image datasets
Facial, OCR, retail and document imagery annotated to your schema.
Learn more
Text datasets
Chat logs, prompt and response pairs, and parallel corpora by domain.
Learn more
Video datasets
Driver monitoring, gesture, 360 vehicle and visual question answering footage.
Learn moreNeed multimodal data we do not already hold?
Tell us the language, dialect, device and environment. We will come back with a recruitment timeline and a price.