This Startup Raised $18 Million to Help Computers Hear You Better
Treble has raised $18 million to grow its business. The funding is an extension of its Series A round and was led by Paladin Capital Group. TechCrunch reported the news with details from the company.
KOMPAS VC, Frumtak Ventures, EIC and Omega ehf also joined the round. Treble has now raised more than $40 million in total. That follows a $12 million investment in 2024.
Treble had earlier called that 2024 round an €11 million Series A, announced on Sept. 3, 2024. The $12 million figure is now used as the official total for that round, with the euro amount kept as background.
Treble is based in Iceland. It was started in 2020 by Finnur Pind and Jesper Pedersen. Both founders are acoustic engineers, experts in how sound travels.
The company builds software that simulates sound and creates artificial audio to train AI systems.
Its tools are made for companies that build AI models, robots and devices like speakers and headsets. The tools make computer-created voice and noise to help AI learn to clean up speech, block background noise and get better at listening. They are for engineers, not everyday listeners.
Treble also tests voice AI in different everyday conditions and gives feedback to research labs. With Hugging Face, it released a shared test for speech recognition in realistic settings. Its named customers include Amazon and Logitech.
The broader context here is simple. Testing sound in real life is slow and hard. It needs special quiet rooms, early versions of devices and many repeat recordings.
In my view, that is why computer simulation is useful. Like a flight simulator for pilots, it lets engineers try many virtual rooms without building them. They can move a device or add kitchen noise with a few clicks. That saves time and makes mistakes easier to find.
Worth flagging is that Treble does two different jobs. One is creation, making artificial voice and noise to add to real recordings. The other is testing, running a voice system through set situations and reporting how it did.
Looking at what this means for builders, the shared test with Hugging Face helps. It gives different teams the same standard to measure against. It does not say which system is best everywhere. It lets teams see what breaks in noisy places before they build hardware or record in the field.
In the long arc, this kind of tool makes good sound cheaper to get right. My children grew up repeating themselves to speakers in loud kitchens and cars. Fixing that is quiet, basic work. It is also what helps people trust voice tools, and testing early helps users.
