Visual Speech Dataset Collection for Emotion Recognition Model

Client's Challenge & Our Solution

An AI company developing emotion recognition models for various industries approached FutureBeeAI to collect a diverse visual speech dataset. The client needed 1000 HD videos featuring participants speaking unscripted monologues on given prompts that showcased a variety of emotions. Additionally, they required diversity in the dataset across participant age, gender, ethnicity, recording device, recording style, background, and the time of day.

FutureBeeAI executed a global, large-scale data collection project, recruiting 250 participants from 10 countries to ensure diverse demographics. We used a participants latest smartphones to capture high-quality videos with varied backgrounds and lighting conditions. The videos were then annotated with emotion labels and carefully reviewed for accuracy.

Outcome & Features:

Collected 1000 HD videos from 250 participants across 10 countries, ensuring all the diversity requirements

Data captured across various devices, recording styles, and times of day to ensure the dataset’s applicability in diverse real-world scenarios.

Completed the project within 8 weeks, providing a robust, real-world multimodal dataset for AI model training