Icelandic startup Treble raises $18M to simulate and test voice AI

The round

Iceland-based startup Treble has raised $18 million to expand its acoustic simulation platform, according to TechCrunch. The funding is an extension of the company's Series A and was led by Paladin Capital Group, with existing backers KOMPAS VC, Frumtak Ventures, EIC and Omega ehf participating.

The round follows a $12 million investment in 2024 and brings Treble's total funding to more than $40 million, TechCrunch reports. The company was founded in 2020 by acoustic engineers Finnur Pind and Jesper Pedersen, and it already counts Amazon and Logitech among its customers.

A simulation layer for voice AI

Voice AI has become one of the most heavily funded corners of the AI market, with investors pouring billions into automated customer support, sales calls, meeting notetakers and smart glasses that use speech as the main way people interact with devices. Treble's bet is that this activity needs a testing and feedback loop, and it wants to be the company supplying that to model developers, robotics firms and consumer hardware makers.

The company works across several connected areas. For voice AI developers, it runs a synthetic data generation platform that can be used for speech enhancement, noise suppression and model training, and it evaluates voice models under varying acoustic conditions to give labs feedback on their performance. Earlier this year, Treble partnered with Hugging Face to launch a benchmark for speech recognition models tested across realistic conditions.

On the hardware side, Treble handles design and testing from an audio perspective. It helps headphone and speaker makers prototype virtually so they can predict how a product will sound before it is built, and it can test how well a smart speaker understands voice commands depending on where the device is placed. More recently it has added simulation testing for smart glasses and other AI devices.

Physics simulation instead of scraped audio

Pind argued to TechCrunch that audio AI is fundamentally a data problem: nearly all sound-related AI to date has been built from recordings and audio scraped from the internet, and Treble believes accurate physics simulation can serve as an alternative way to produce training data for sound.

That thesis is what drew in the new lead investor. Francois Ruether, a vice president at Paladin Capital Group, told TechCrunch that as more products depend on interpreting sound, Treble's infrastructure grows more valuable across voice AI, wearables, robotics and physical AI. He added that customers keep ownership of their own models, products and development workflows while drawing on a shared, simulation-based acoustic infrastructure layer.

Expansion into physical AI

Treble plans to use the fresh capital to push deeper into physical AI, targeting robotics, automotive and drone companies that want to add sound-based functions through testing and simulation.

Pind also pointed to hearing-enhancing wearables as an area of particular interest, describing a future generation of headphones and smart glasses that could let users hear better in difficult acoustic settings, such as focusing only on people within a couple of meters at a restaurant or muting surrounding conversation during a seminar.

Why it matters

Most of the attention in voice AI goes to the labs building speech models and the hardware companies shipping voice-first devices. Treble is chasing the layer underneath both: the data and evaluation infrastructure that determines whether these systems actually work in messy, real-world acoustics. If physics-based simulation can credibly generate training data and reproduce difficult listening conditions, it addresses two persistent bottlenecks at once, namely the scarcity of high-quality audio data and the difficulty of testing devices before physical prototypes exist. The raise is also a signal that investors see durable value in picks-and-shovels tooling for voice and physical AI rather than only in the frontier models themselves.