Iceland-based Treble raises $18 million for its voice simulation platform

Voice AI has emerged arsenic 1 of the hottest sectors successful AI, with investors pouring successful billions of dollars to service usage cases ranging from automating lawsuit enactment and income calls to creating gathering notetakers and processing AI astute glasses that usage dependable arsenic the superior enactment surface.

Meanwhile, AI labs are rapidly releasing models, and hardware makers are trying to make the champion acquisition for consumers to interact with devices.

All of this caller exertion needs investigating and a feedback loop for improvement. An Iceland-based startup, Treble, is positioning itself to beryllium astatine the halfway of the dependable AI manufacture by creating a simulation level that tin cater to exemplary makers, robotics companies, and user hardware makers.

The company, founded successful 2020 by acoustic engineers Finnur Pind and Jesper Pedersen, has raised $18 cardinal successful an hold of its Series A backing led by Paladin Capital Group, with information from existing investors KOMPAS VC, Frumtak Ventures, EIC, and Omega ehf. It received an concern of $12 cardinal successful 2024, bringing its full rise to day to implicit $40 million. The startup counts Amazon and Logitech arsenic customers.

Treble co-founder Finnur Pind Image Credits: Treble Image Credits:Treble

Treble has a fewer verticals related to simulation and data. For dependable AI companies, it has a synthetic information procreation level that could beryllium utilized for code enhancement, sound suppression, and exemplary training. It besides evaluates dependable AI models successful antithetic conditions to supply feedback to labs. Earlier this year, it partnered with Hugging Face to motorboat a benchmark for code designation models across antithetic realistic conditions.

“Audio AI is truly a information challenge, and this is wherever the astir opportunities to alteration next-generation models and hardware lie. To date, beauteous overmuch each sound-related AI has been made from recordings and information scraped from the internet. We judge that close physics simulation tin beryllium an alternate mode to make information for sound,” Pind told TechCrunch.

The startup besides focuses connected hardware plan and investigating from a dependable perspective. For instance, it works with headphone and talker companies for virtual prototyping to assistance them recognize however their merchandise mightiness sound. It could besides trial however a astute talker tin recognize commands based connected the positioning of the speaker. Lately, Treble has ventured into providing simulation investigating for astute glasses and AI devices.

Pind said that the institution is excited to enactment connected wearables that could heighten proceeding for users.

“I’m truly excited astir the adjacent procreation of these devices similar headphones and astute glasses that tin alteration [a diagnostic like] superhuman hearing. That’s an country wherever you tin truly conscionable perceive amended successful challenging acoustic environments. Maybe you are successful a restaurant, and you lone privation to perceive radical wrong 2 meters of range, oregon you are successful a seminar, and privation to mute radical astir you,” helium said.

Treble aims to summation its absorption successful the carnal AI space, including robotics, automotive, and drone companies, to alteration sound-based functions for them done investigating and simulation.

Francois Ruether, VP of Paladin Capital Group, told TechCrunch that Treble’s platform, which simulates antithetic models and devices, stands out; the platform’s value volition summation arsenic it gets progressive successful much areas.

“Our thesis is that, arsenic much products beryllium connected knowing sound, this infrastructure becomes progressively invaluable crossed dependable AI, wearables, robotics, and carnal AI. Customers clasp ownership of their models, products, and improvement workflows, portion benefiting from a shared instauration of a simulation-native acoustic infrastructure layer,” Ruether said.