TL;DR: We provide high-quality human voice and video datasets for companies building realistic AI models. Real data sourced in-house across diverse languages, emotions, and contexts.

Ask: Talk to us if you're a researcher or founder building voice/video generation models. We can help you.

https://youtu.be/PwjTSZRC8XE

The Problem:
Companies building voice and video AI models need quality training data to reach human-like performance for models. Scraping has limits and existing datasets lack diversity across accents, emotions, and languages. As AI moves toward human-like interactions like customer conversations, therapy sessions, classroom teaching, or entertainment, models need authentic human expressions that the internet can't supply.

Our Solution:
Liva collects real, consented human voice and video entirely in-house—no synthetic data or third-party purchases. All content is authentic and rights-cleared.  We’re already delivering our voice dataset to a lab training expressive foundation models for voice.

We capture diverse accents, emotional range, and varied contexts (sales calls, multi-channel dialogues, expressive monologues, casual conversations, job interviews, and more) with high production quality through crowdsourcing and strategic partnerships.

The Team:

We’ve worked on many research projects together since we met 3 years ago.

• Ashley: Caltech CS dropout. Built an AI model to detect lung disease from cough recordings (@ MIT) and led large-scale data collection initiatives. 
• Aoi: Prev. Harvard Bio/CS. Did ML research in representation learning and image diffusion models. 5+ Publications in ICML, Nature, etc.

Our Ask:

Introductions to:

• AI labs building or fine-tuning voice, video, and multimodal generation models
• Companies using or integrating voice/video models
• Film, audio, and content production companies
• Researchers working on audio, video, and multimodal generative models

Contact us: founders@theliva.ai