Deepslate, a Brandenburg-based AI model lab headquartered in Templin, has secured €7.7 million in Seed funding to advance speech-to-speech AI models built and hosted entirely in Europe. The round was led by 42CAP, with participation from Alstin Capital, existing investor SIVentures, formerly Smart Infrastructure Ventures, and several business angels.
The company, whose leadership includes Paskal Paesler and Jan Brachthäuser, develops end-to-end models that process spoken language directly rather than converting it into text first. Deepslate said this approach reduces latency while preserving tone, emphasis and dialect. Its system combines an in-house speech encoder, a reasoning core built on an open-weights language model post-trained for European languages, and a speech decoder that generates the spoken reply.
According to the company, its model recorded a 440-millisecond response time in an independent Artificial Analysis benchmark, the fastest speech-to-speech result measured as of September 2026, and posted the best error rate for European languages in the CoVoST2 comparison. Insurers, contact centers and platforms already use the technology in production.
Deepslate said European data sovereignty is a core requirement for its customers, with all hosting inside the EU and self-hosting available. The new capital will fund expanded model training, a German-language data program covering street names, personal names and dialects, and growth in sales, marketing and European production infrastructure.
Featured image: Credit: Deepslate
