Treble, an Iceland-based acoustic simulation startup, has raised $18 million in a Series A extension led by Paladin Capital Group, pushing its total funding above $40 million. The round follows a $12 million investment in 2024 and lands as voice AI becomes one of the most heavily funded sub-sectors in the industry, spanning customer support agents, meeting notetakers, and AI-first smart glasses. KOMPAS VC, Frumtak Ventures, EIC, and Omega ehf all returned as investors.
The company sells physics-based acoustic simulation as infrastructure to two distinct buyers: AI labs training speech models, and hardware companies designing devices that emit or capture sound. Amazon and Logitech are named customers. Treble was founded in 2020 by acoustic engineers Finnur Pind and Jesper Pedersen.
The pitch is that most speech and audio models today are trained on internet-scraped recordings, which limits how well they generalize to unusual rooms, noisy restaurants, or the specific microphone arrays inside a new pair of glasses. Treble generates synthetic acoustic data by simulating how sound waves actually behave in a modeled physical space, then feeds that data into training pipelines for speech enhancement, noise suppression, and model evaluation.
“Audio AI is really a data challenge, and this is where the most opportunities to enable next-generation models and hardware lie. To date, pretty much all sound-related AI has been made from recordings and data scraped from the internet.”— Finnur Pind, Treble co-founder
Key facts
- 01Treble raised $18M in a Series A extension led by Paladin Capital Group, bringing total funding above $40M.
- 02Existing backers KOMPAS VC, Frumtak Ventures, EIC, and Omega ehf joined the round; the last raise was $12M in 2024.
- 03Amazon and Logitech are named customers of Treble's acoustic simulation platform.
- 04Treble partnered with Hugging Face earlier this year to launch a speech-recognition benchmark across realistic acoustic conditions.
- 05Founded in 2020 by acoustic engineers Finnur Pind and Jesper Pedersen, the company is based in Iceland.
The company has built out a benchmarking arm alongside the data-generation product. Earlier this year, Treble partnered with Hugging Face to launch a benchmark that scores speech recognition models across a range of realistic acoustic conditions, giving labs a standardized way to see where their systems break.
On the hardware side, Treble runs virtual prototyping for headphone and speaker companies, letting them hear how a design will sound before a physical unit exists. The same simulation stack tests how well a smart speaker picks up commands from different positions in a room, and Pind said the company is now applying it to smart glasses and AI-first wearables.
Pind pointed to hearing enhancement as one of the more interesting near-term applications: glasses or headphones that let a wearer isolate voices within a two-meter range in a loud restaurant, or mute background chatter during a seminar. Building and tuning those features requires enormous volumes of acoustic scenarios that are impractical to record in the physical world.
“Our thesis is that, as more products depend on understanding sound, this infrastructure becomes increasingly valuable across voice AI, wearables, robotics, and physical AI.”— Francois Ruether, VP of Paladin Capital Group
Treble is also expanding into what it calls the physical AI space, targeting robotics, automotive, and drone customers that need machines to interpret sound reliably in unstructured environments. That market has grown quickly as robotics companies pair large vision models with audio channels for spatial awareness and human interaction.
The counterweight to Treble's story is competition and category risk. Synthetic data for voice is not an uncontested lane — several well-funded voice AI companies build their own data pipelines in-house, and hyperscaler cloud providers have their own tooling for speech model evaluation. Treble's bet is that a dedicated simulation-native layer is easier to adopt than reinventing physics inside every customer's stack.
For an $18M extension out of Iceland, the customer roster is the tell. Amazon and Logitech are not exploratory design partners — they are shipping-scale hardware companies with mature audio teams, and they are paying for simulation rather than building it. That signals the emerging voice-AI infrastructure layer is going to look a lot like the picks-and-shovels layer of the LLM boom: horizontal tooling companies quietly selling into every model lab, wearables vendor, and robotics startup that needs sound to just work.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
