Voice AI has emerged as one of many hottest sectors in AI, with buyers pouring in billions of {dollars} to serve use instances starting from automating buyer help and gross sales calls to creating assembly notetakers and growing AI good glasses that use voice as the first interplay floor.
In the meantime, AI labs are rapidly releasing fashions, and {hardware} makers are attempting to create the most effective expertise for customers to work together with units.
All of this new expertise wants testing and a suggestions loop for enchancment. An Iceland-based startup, Treble, is positioning itself to be on the heart of the voice AI business by making a simulation platform that may cater to mannequin makers, robotics corporations, and shopper {hardware} makers.
The corporate, based in 2020 by acoustic engineers Finnur Pind and Jesper Pedersen, has raised $18 million in an extension of its Collection A funding led by Paladin Capital Group, with participation from present buyers KOMPAS VC, Frumtak Ventures, EIC, and Omega ehf. It obtained an funding of $12 million in 2024, bringing its whole increase so far to over $40 million. The startup counts Amazon and Logitech as clients.
Treble has a couple of verticals associated to simulation and knowledge. For voice AI corporations, it has an artificial knowledge era platform that could possibly be used for speech enhancement, noise suppression, and mannequin coaching. It additionally evaluates voice AI fashions in several circumstances to supply suggestions to labs. Earlier this 12 months, it partnered with Hugging Face to launch a benchmark for speech recognition fashions across different realistic conditions.
“Audio AI is known as a knowledge problem, and that is the place essentially the most alternatives to allow next-generation fashions and {hardware} lie. So far, just about all sound-related AI has been created from recordings and knowledge scraped from the web. We imagine that correct physics simulation could be another option to create knowledge for sound,” Pind instructed TechCrunch.
The startup additionally focuses on {hardware} design and testing from a voice perspective. As an illustration, it really works with headphone and speaker corporations for digital prototyping to assist them perceive how their product would possibly sound. It may additionally check how a wise speaker can perceive instructions based mostly on the positioning of the speaker. Currently, Treble has ventured into offering simulation testing for good glasses and AI units.
Pind stated that the corporate is happy to work on wearables that might improve listening to for customers.
“I’m actually excited concerning the subsequent era of those units like headphones and good glasses that may allow [a feature like] superhuman listening to. That’s an space the place you may actually simply hear higher in difficult acoustic environments. Perhaps you’re in a restaurant, and also you solely need to hear folks inside two meters of vary, or you’re in a seminar, and need to mute folks round you,” he stated.
Treble goals to extend its focus within the bodily AI area, together with robotics, automotive, and drone corporations, to allow sound-based capabilities for them by testing and simulation.
Francois Ruether, VP of Paladin Capital Group, instructed TechCrunch that Treble’s platform, which simulates completely different fashions and units, stands out; the platform’s significance will improve because it will get concerned in additional areas.
“Our thesis is that, as extra merchandise rely upon understanding sound, this infrastructure turns into more and more priceless throughout voice AI, wearables, robotics, and bodily AI. Clients retain possession of their fashions, merchandise, and growth workflows, whereas benefiting from a shared basis of a simulation-native acoustic infrastructure layer,” Ruether stated.
Whenever you buy by hyperlinks in our articles, we may earn a small commission. This doesn’t have an effect on our editorial independence.