Image Audio Description Dataset Creation

Client's Challenge & Our Solution

A leading tech company developing multimodal AI systems sought our expertise to create a dataset of unscripted image audio descriptions. The objective was to have native language experts describe images provided to them, capturing both the essence and details of each image in natural language.

FutureBeeAI utilized our Yugo platform to manage the entire data collection process, ensuring streamlined project tracking and quality assurance. We onboarded over 500 native language experts fluent in rare Indian languages such as Marwadi, Kutchhi, Sindhi, Goumukhi, and Sanskrit, along with major Indian languages like Hindi, Tamil, Telugu, Gujarati, and more. This diverse linguistic expertise allowed us to collect high-quality, unscripted audio descriptions in over 15 languages, including rare dialects, delivering the dataset within 8 weeks.

Outcome & Features:

Successfully recorded unscripted image audio descriptions with 500 native language experts on 10,000 images for each language

Collected dataset across 15 rare and major indian languages

Multimodal dataset collected within 8 weeks with FutureBeeAI crowd community and Yugo platform