Training a voice model usually means feeding it recordings scraped off the open internet. Treble’s bet is that physics can generate better data than the web can supply.
The Reykjavik company was founded in 2020 by two acoustic engineers, Finnur Pind and Jesper Pedersen. It has now added $18M in an extension of its Series A, led by Paladin Capital Group with KOMPAS VC, Frumtak Ventures, EIC and Omega ehf taking part. A $12M investment in 2024 had already pushed the lifetime total past $40M.
Simulation is the product, not a model. Treble argues that audio AI is at heart a data problem, so it generates synthetic material for speech enhancement, noise suppression and training, and it runs voice models through varied acoustic conditions to return findings to the labs. One earlier collaboration with Hugging Face produced a speech recognition benchmark built around realistic environments. Amazon and Logitech are among its customers.
The other half of the business is hardware. Speaker and headphone makers use Treble to hear a design before it exists, and the company measures how a smart speaker reacts to commands depending on where in a room it stands.
Pind is most interested in what comes next for wearables. He describes headphones and smart glasses that let a wearer in a restaurant pick out only voices within two meters, or silence a crowd during a seminar.
Robotics, automotive and drones form the next target. Paladin vice president Francois Ruether said the reasoning is simple: as more products depend on hearing, an acoustic simulation layer becomes more valuable, and customers keep their own models and workflows.
