Research

State-of-the-art text & speech models trained for real-time interactivity

We train our own tiny text and speech models. The architecture of these models and our training pipelines are optimized for uniqueness. We want to help anyone that we work with train models that are uniquely theirs and give them a creative edge.

Real-time & performant

Our model architecture is optimized for interactive, streaming use cases.

Memory efficient

Runs comfortably on any consumer device, including low-end phones.

Creative differentiation

Our training pipelines are designed to produce maximally differentiated models.