The Storytelling Video Dataset offers a rich collection of high-quality video recordings featuring 700 native Russian speakers. Each participant shares unscripted narratives for over 10 minutes, capturing both full-body movements and an array of non-verbal cues such as gestures and facial expressions. This unique dataset is specifically designed to facilitate the training and development of multimodal AI models.

Dataset Overview

• Participants: 700 unique individuals
• Video Duration: 10+ minutes per participant
• Quality: At least Full HD (1920x1080), with many samples available in 2K and 4K, recorded at 30 FPS
• Content: Natural, unscripted speech in Russian
• Framing: Full-body view (waist-up or lower)
• Features: Clear visibility of facial expressions, gestures, and body language along with clean lighting and audio
• Commercial Use: Rights secured for commercial applications
• Metadata: Includes age and gender information of participants

Use Cases

The dataset is ideal for applications in several advanced fields:

• Multimodal model training
• Emotion and gesture recognition
• Conversational avatar creation
• VR/AR training simulations
• Natural Language Processing (NLP) and speech-to-text applications

Additional Resources

• For a glimpse of the data structure, refer to the sample video dataset.
• Find ample visual references in the provided screenshot gallery.
• For detailed licensing information, consult the license agreement.

This dataset stands out in its ability to combine visual and auditory information, making it a robust resource for anyone engaged in AI training focusing on linguistic behavior and non-verbal communication.

GitHub: github.com/MaratDV/storytelling-video-dataset

Hugging Face: huggingface.co/datasets/MaratDV/video-dataset

License: Commercial use only — contact required

Contact: chinzad@gmail.com | Telegram: @Marat_DV