Audio · Video · Multimodal / AI & Machine Learning
Audio-Video Conversational Dataset | 4,000 Hours of Global Synchronized Speech & Video
Advertised by providerLicensing rights not independently verifiedPrice on request
Overview
Audio-Video Conversational Dataset | 4,000 Hours of Global Synchronized Speech & Video is a multimodal dataset. The provider states a volume of 4,000 hours. Typical applications include speech recognition and conversational AI. Relevant sectors: AI & Machine Learning and Education. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Stated volume of 4,000 hours.
The Audio-Video Conversational Dataset is a 4,000-hour collection of synchronized speech and video recordings of people having natural conversations across diverse speakers, languages, accents, environments, and interaction styles. The dataset is designed for multimodal voice agents, avatar systems, lip-sync models, emotion recognition, paralinguistic understanding, conversational video models, audio-video MLLM training, and human interaction world modeling. Each recording pairs synchronized conversational audio with video that captures face movement, mouth motion, gestures, eye gaze, facial expressions, pauses, laughter, interruptions, overlapping speech, and turn-taking behavior. This structure helps models learn how speech, expression, timing, and body language interact during real human communication. Language coverage includes English, Spanish, Portuguese, Mandarin Chinese, Hindi, Arabic, French, German, Japanese, Korean, Indonesian, Vietnamese, Thai, Turkish, Tagalog, Italian, Dutch, Malay, Urdu, Tamil, Telugu, and additional regional or underrepresented languages. The language distribution is structured for global model training, with approximately 25% English and major English accents, 15% Spanish and Portuguese, 15% East Asian languages, 12% South Asian languages, 10% European languages, 8% Arabic and MENA regional speech, 10% Southeast Asian languages, and 5% long-tail or underrepresented languages. Technical specifications include 4,000 hours of synchronized audio-
Related datasets
More in Video & Physical AI →Audio · Video · MultimodalAdvertised
Speech Data | 20,000+ Hours of Real Job Interview Audio & Video | Worldwide | Consented & Unscripted
What this dataset is the provider licenses real job interview recordings captured on its own AI interviewer platform, used by 4,000+ companies.
- Language
- —
- Volume
- 20k h
- Domain
- AI & Machine Learning +1
- Use case
- Speech recognition
Price on request