A different scale.
We’re an independent research lab working on small audio models, starting with speech.
What can we do
with a little less?
Less compute. Less memory. A model that more people can run, study, and build on. Those constraints shape the questions we want to ask.
Our first focus is Rivet, our core text-to-speech project. We’re interested in the details that make speech work: intelligibility, rhythm, expression, and the practical cost of generating it.
Meet RivetMake constraints useful.
Study the tradeoffs behind smaller models. Report the hardware and settings alongside the performance.
Share the whole finding.
Make room for careful measurements, difficult examples, limitations, and negative results.
Leave a path to reproduce it.
Publish enough context for someone else to understand the result and test the assumptions.
Inspired by open science.
Kyutai’s approach to making research accessible through publications, tools, and open releases is a deep inspiration for us. We want to contribute to that culture at our own scale.
Missing Fundamental Labs is independent of Kyutai. We’ll credit the ideas, tools, and models we build on throughout our research.
Explore KyutaiA note can be heard
even when it isn’t there.
The missing fundamental is an auditory phenomenon: remove a sound’s fundamental frequency, and its harmonics can still evoke the same pitch. A small absence, a surprisingly complete perception.
The story behind our name