Paladin leads $18 million into Treble's simulated rooms for testing voice AI

The bug report a user files is that the speaker did not hear them from the sink. Treble sells a test for that. Its platform can check how well a smart speaker understands commands depending on where the speaker is positioned ^[15]. Run that in software and the failing case has a room in it, and you can run it again after the industrial designer moves the microphone.

Subtract the $12 million Treble took in 2024 and the $18 million announced now from a disclosed total above $40 million, and at least $10 million predates both rounds ^[21].

Finnur Pind's pitch is about method. "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. We believe that accurate physics simulation can be an alternative way to create data for sound," Pind told TechCrunch ^[6]. A team that trains noise suppression on simulated rooms takes on the difference between the simulated room and the customer's kitchen. Treble also evaluates voice models under different conditions and feeds the results back to labs ^[4].

Francois Ruether, VP of Paladin Capital Group, told TechCrunch that the platform stands out and will matter more as it reaches more areas ^[9]. "Our thesis is that, as more products depend on understanding sound, this infrastructure becomes increasingly valuable across voice AI, wearables, robotics, and physical AI. Customers retain ownership of their models, products, and development workflows, while benefiting from a shared foundation of a simulation-native acoustic infrastructure layer," Ruether said ^[7]. Ruether is an investor describing a layer he has just bought into. TechCrunch's report does not give Treble's pricing, a customer count beyond Amazon and Logitech ^[17], or any example of a buyer requiring acoustic test coverage from a supplier.

The customer for this is a model maker, a robotics company, or a consumer hardware maker ^[13]. A team wiring a support-call agent onto someone else's speech API gets that coverage second-hand, if at all.

For anyone shipping a voice feature next month, two things separate one failure from another: whether a user will report it or only your logs see it, and whether you can reproduce the condition on demand. Simulation is what lets you reproduce it on demand. It can tell you a model holds up in a modelled restaurant; knowing how many of your users are in restaurants still comes from recordings of real ones. Treble has moved into testing smart glasses and AI devices ^[16], and Pind said he is excited about a next generation of devices such as headphones and smart glasses that could deliver superhuman hearing, meaning hearing better in challenging acoustic environments: a restaurant where you only want to hear people within two meters of range, or a seminar where you want to mute the people around you ^[19].