A voice agent can sound almost human until a caller says “Müller-Straße 14b,” reads an IBAN aloud, switches from German to English halfway through a sentence or speaks in a regional dialect. For Deepslate, those are not edge cases. They are the reason to build a European voice model in the first place.
The German voice AI company has raised €7.7 million in seed funding, led by 42CAP with participation from Alstin Capital, existing investor SIVentures and several business angels. The capital will go toward model research, European training data, commercial expansion and additional infrastructure in European data centres.
At the centre is Opal, Deepslate’s proprietary speech-to-speech model. Conventional voice systems often move through three stages: speech recognition turns audio into text, an LLM reasons over that text, and text-to-speech converts the answer back into audio. Opal instead operates directly on speech representations, which Deepslate says helps preserve tone, emphasis and other acoustic information while avoiding errors that can accumulate between separate systems.
That architecture becomes more interesting when paired with Deepslate’s European focus. The model supports 27 languages, while the company is training specifically around names, addresses, dialects and structured information such as phone numbers, policy references and meter readings. Its sector pages position the same underlying voice layer across contact centres, insurance, utilities, telecoms, healthcare and public services.
Speed is part of the argument too. Deepslate currently cites a 440-millisecond time-to-first-audio result from Artificial Analysis in September 2026, while its technical documentation separately describes sub-300ms first-byte latency. Those are different measurements, but both point to the same product priority: making conversational pauses less noticeable.
The second half of Deepslate’s pitch is sovereignty. Its cloud infrastructure is located in Europe, the company is ISO 27001 certified, and enterprise customers can self-host the complete voice stack, including in isolated environments where audio never leaves company infrastructure.
Deepslate is already being used in production by insurers, contact centres and platform providers. Founders Paskal Paesler and Jan Brachthäuser are now using the seed round to push the technology deeper into European enterprise infrastructure.
The larger opportunity is not simply teaching AI to speak naturally. It is making sure that when Europe speaks to AI, the model understands what was actually said, where that conversation can legally go and how quickly the answer needs to come back.
