Deepslate Secures €7.7M Seed Funding to Build European Speech-to-Speech AI Models

Deepslate is an AI model lab headquartered in Templin, Brandenburg, Germany. It has raised a €7.7 million seed round to advance a speech-to-speech model 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 angel investors. The company's leadership includes Paskal Paesler and Jan Brachthäuser.

The story this money is meant to tell has two pillars: the technical approach and data sovereignty. The former concerns performance; the latter concerns why customers must choose a European vendor. Together they form the narrative skeleton of this round, but the evidence supporting them is not equally strong.

Deepslate develops end-to-end models that process spoken language directly, rather than converting speech to text first and then processing it. Its system consists of three parts: a self-developed speech encoder, an inference core post-trained on an open-weight language model and tailored to European languages, and a speech decoder that generates spoken replies.

The company says this direct processing of speech can reduce latency while preserving tone, stress and dialect. This causal judgment is a company claim; the report provides no controlled experiments or ablation data to prove that "direct processing" itself delivers the latency advantage. What can be confirmed is the architecture description; what cannot be confirmed is the causal chain between the architecture and performance.

According to Deepslate, its model recorded a 440-millisecond response time on the independent Artificial Analysis benchmark, the fastest speech-to-speech result measured as of September 2026; it also says it achieved the best error rate for European languages in a CoVoST2 comparison.

These two numbers are the most shareable part of the entire report, but both must be labeled as company claims. The report does not provide the full methodology or raw data of the benchmark tests, nor does it specify the exact test conditions, model version or comparison targets of the "independent Artificial Analysis benchmark." Therefore, whether the scope of the "fastest" and "best" conclusions is complete and whether they can be independently reproduced cannot currently be verified.

A distinction is needed: the funding facts, headquarters location, leadership, architecture composition and use of funds are confirmable information; the performance rankings are company claims. Citing the latter as established industry fact would exceed what the source can support.

Deepslate says European data sovereignty is a core requirement for its customers, that all hosting is within the EU, and that a self-hosting option is available. This is the company's statement about its customers' motivations; the report provides no customer interviews or contract details to corroborate it.

The report also does not explain the actual compliance status and certifications of the EU hosting and self-hosting options. Therefore, "data sovereignty is a core requirement" can be understood as a market positioning, but not as a third-party-verified industry conclusion.

The report says insurers, call centers and platforms are already using the technology in their production environments. This claim likewise comes from the company, and the report provides no specific names of production customers or deployment scale.

In the absence of a customer list or third-party confirmation, this information can only be treated as a company claim. It shows the company wants to signal "commercial deployment already exists," but it cannot be used to judge revenue scale or customer concentration.

The new funds will be used to scale model training, carry out a German-language data project covering street names, personal names and dialects, and expand sales, marketing and European production infrastructure. This is the more specific, confirmable description of use in this round.

However, the report does not explain the scale, timeline or data-source compliance details of the German-language data project, nor does it disclose the valuation, total round size or timetable for the use of funds. The direction of the use of funds is clear, but the execution pace and resource allocation remain opaque.

The above information on funding, technology and benchmark performance all comes from the same report, with no other independent sources provided. This means that even if the funding facts themselves are fairly credible, the technical performance and customer deployment parts still lack cross-verification.

In addition, the technology's performance beyond European languages, model size and degree of open-sourcing were not disclosed. For a lab positioning itself around "European languages" as its differentiation, these gaps are precisely the key variables for assessing its long-term competitiveness.

Therefore, the more prudent reading is: this is an early-stage round with a clear direction, and the company has given a clear architectural narrative and a set of eye-catching but not independently reproduced performance claims. The funding is confirmed; the performance awaits verification.