Developing Multilingual Visual Speech Datasets to Enhance Emotion Recognition

Client's Challenge & Our Solution

A leading AI company approached FutureBeeAI to collect a unique multilingual visual speech dataset to train their emotion recognition model. The project involved capturing unscripted speech responses in multiple languages, where participants answered prompts while showcasing diverse emotions such as happiness, sadness, excitement, shock, and neutral tones.

FutureBeeAI leveraged its global crowd of participants to provide high-quality video and audio recordings, ensuring a wide range of visual and auditory cues across different emotional states to improve the AI's ability to recognize and respond to emotional cues. The dataset was delivered along with visual speech data, extensive metadata, and transcription.

Outcome & Features:

Collected 1000+ high-resolution visual speech videos in multiple languages, showcasing emotions like happy, sad, excited, and neutral.

Ensured the inclusion of diverse demographics and cultural nuances in emotional expression for broader model applicability.

Delivered a fully structured dataset with proper metadata and naming that significantly enhanced the client’s emotion recognition model.