Iceland’s Treble raises $18 million to test the next generation of voice AI hardware

Iceland-based Treble has raised $18 million in an extension of its Series A to expand a platform that simulates real-world acoustic conditions for voice AI developers, hardware makers, and robotics companies. The funding gives the startup more capital to develop testing and synthetic-data tools as speech interfaces move from software into devices such as smart glasses, headphones, and robots.

Paladin Capital Group led the round, with participation from existing investors KOMPAS VC, Frumtak Ventures, the European Innovation Council, and Omega ehf, according to TechCrunch. Treble raised $12 million in 2024, bringing its total funding to more than $40 million. The company counts Amazon and Logitech among its customers.

Funding follows a broader voice AI testing problem

The round arrives as voice AI companies release models faster and device manufacturers experiment with voice as a primary interface. Those systems must work in environments that are substantially less controlled than a benchmark dataset: crowded restaurants, moving vehicles, homes with multiple speakers, or rooms where a device is positioned away from the user.

Treble, founded in 2020 by acoustic engineers Finnur Pind and Jesper Pedersen, is building infrastructure for that testing layer. Its voice simulation platform is designed to reproduce different acoustic environments and device configurations, allowing teams to assess how models and products perform before extensive physical testing.

That focus distinguishes the company from tools aimed only at generating or evaluating text and speech in clean conditions. For builders of voice interfaces, failures often come from background noise, reverberation, speaker placement, competing voices, or microphones that behave differently across hardware designs. Simulating those conditions can make testing more repeatable, although the available evidence does not establish how closely Treble’s simulations match every real-world environment.

From synthetic data to device design

According to TechCrunch, Treble offers synthetic data generation for use cases including speech enhancement, noise suppression, and model training. The company also evaluates voice AI models under varied conditions and provides feedback to developers.

The startup’s approach is based on physics-based acoustic simulation rather than relying exclusively on recordings or audio scraped from the internet. Pind told TechCrunch that sound-related AI remains fundamentally a data challenge and argued that accurate simulation can provide an alternative way to create training and evaluation data.

Treble is also applying the technology to hardware development. The company works with headphone and speaker makers on virtual prototyping, including analysis of how a product may sound and how speaker placement affects command recognition. That could help product teams identify microphone and positioning problems earlier, before committing to multiple physical prototypes.

The company has recently extended this work to smart glasses and other AI devices. Pind also pointed to hearing-enhancement applications, such as isolating nearby speech in a noisy restaurant or reducing surrounding sound during a seminar. Those examples are potential product directions, not evidence that Treble has launched a consumer feature or that the underlying devices can already deliver those results.

What the evidence says about adoption and performance

The confirmed funding and customer information comes from TechCrunch’s report, which identified Amazon and Logitech as Treble customers. The source did not disclose contract values, revenue, customer counts, deployment scale, or the specific products using Treble’s platform.

Earlier in 2026, Treble partnered with Hugging Face to launch a benchmark for speech recognition models across realistic conditions, according to the same report. The benchmark indicates an effort to make speech evaluation more representative of deployment environments, but the available source material does not provide scores, methodology details, or evidence that it has become a widely adopted industry standard.

The investor case was described by Francois Ruether, vice president of Paladin Capital Group. He said Treble’s ability to simulate different models and devices could become more valuable as products depend on understanding sound across voice AI, wearables, robotics, and physical AI. That is an investor thesis rather than an independently verified market forecast.

There are also no reported comparative results showing that Treble’s simulations improve model accuracy, reduce testing costs, or shorten hardware development cycles by a specific amount. For enterprise buyers, those metrics will matter as much as the breadth of the platform.

Why the round matters to builders and enterprises

For AI model teams, Treble’s tools could support controlled testing across noise profiles and room conditions before a model is shipped. This may be particularly useful for teams building voice assistants, meeting tools, or embedded speech systems that must operate across many microphones and environments. Synthetic data could also help fill gaps where collecting and labeling enough real-world audio is expensive, slow, or difficult to govern.

Hardware teams face a different challenge. Smart glasses, headphones, smart speakers, and robots combine microphones, speakers, processors, and physical enclosures, so a change in industrial design can affect recognition quality. Virtual prototyping may allow engineers to compare configurations earlier and reduce dependence on repeated physical builds.

The limitations are equally important. Simulations are models of the physical world, not the physical world itself. Unusual reflections, human behavior, device defects, dialect variation, and unexpected background sounds may still be missed. Buyers will need to validate simulated results against field recordings and live testing, especially for safety-sensitive robotics, automotive systems, and assistive hearing products.

Treble’s next challenge is therefore not only expanding its customer base. It must show that a simulation-native acoustic layer produces reliable decisions for model training, product design, and deployment monitoring. It will also compete for budgets that may currently sit with internal evaluation teams, data-collection programs, hardware labs, or general-purpose testing platforms.

What to watch next

The clearest follow-up signals will be Treble’s expansion into robotics, automotive systems, and drones, areas the company says it wants to prioritize. Those markets could test whether its platform can support sound-based functions beyond consumer voice interfaces.

Builders should also watch for public details about the Hugging Face benchmark, including evaluation conditions, participating models, and whether results correlate with performance in deployed products. Customer disclosures from Amazon, Logitech, or additional hardware companies would provide more evidence of commercial adoption.

Finally, Treble’s future announcements should clarify how the new capital will be allocated, whether the platform supports continuous testing during deployment, and how customers measure return on investment. Revenue growth, repeat usage, disclosed accuracy improvements, and reductions in physical prototyping would be more informative than broad claims about the growth of voice AI.

Creati.ai perspective

Treble is targeting an underdeveloped but important layer of the voice AI stack: the gap between a model that performs well in a controlled evaluation and a product that must understand speech in a messy physical environment. Its combination of synthetic data, model evaluation, and hardware simulation gives the company a coherent position across AI wearables, speakers, and robotics.

The funding validates investor interest in acoustic simulation, but it does not yet prove that simulation can replace enough real-world data or testing to become indispensable infrastructure. Treble’s progress will depend on transparent benchmarks, repeatable customer outcomes, and evidence that its tools improve product decisions rather than simply add another testing dashboard.