Deepslate’s €7.7M Seed Round Backs a European Speech-to-Speech AI Model

Berlin-based Deepslate has raised €7.7 million in seed funding to develop its speech-to-speech voice AI. The round was led by 42CAP, with Alstin Capital, existing investor SIVentures and several business angels also participating.25 The company’s pitch is a model that handles spoken input and spoken output directly, alongside EU hosting and options for customers to run it themselves.

How Deepslate’s speech-to-speech model works

Deepslate’s Opal model is designed to process audio input and generate audio output within a single speech-to-speech system. That differs from a common voice-AI setup that passes a conversation between speech recognition, a language model and text-to-speech components. Deepslate says its model handles speech directly rather than first converting it into text.215

The distinction is architectural, not proof on its own that every conversation will feel faster or more natural. Deepslate’s stated aim is to avoid the added steps of a connected pipeline; the experience will still depend on how the model performs in a particular product and deployment.

Performance and language claims

Deepslate advertises 440-millisecond latency and support for 27 languages, with a focus on European languages, accents and dialects.618 A report citing Artificial Analysis also gives a 0.44-second response-time figure, but says that result placed Deepslate just behind Krafton’s Raon model.6 Another report repeats a claim that Deepslate is the fastest of its kind.4 That difference is a reason to treat “fastest” as a reported claim rather than a settled, context-free ranking.

Latency figures and language counts do not tell the whole story of real-world quality. The available reporting does not establish that the advertised speed or language performance will be identical across every language, use case or customer setup.

EU hosting, self-hosting and data sovereignty

Deepslate says its technology is hosted in the EU and that it runs servers in Germany. It also presents self-hosting as an option for organisations that want to keep the system within their own infrastructure.418 The company describes its service as GDPR-compliant, but that statement is a company claim; the available sources do not independently establish a certification or assess a specific customer deployment.18

For buyers, the distinction between EU cloud hosting and self-hosting matters: the first uses the provider’s managed infrastructure, while the second gives the customer more control over where the system runs. Organisations still need to evaluate the configuration and contractual terms against their own data requirements.

How customers can use the technology

Deepslate markets its speech-to-speech technology to builders and businesses, including through an API and self-hosting options.18 The model can serve as the voice layer in a conversational product; how it connects to a customer’s other software depends on that implementation. The available sources describe intended uses across areas such as healthcare, energy, telecommunications and public services, but do not independently verify deployments or outcomes in those sectors.18

What Deepslate plans to do with the funding

The company plans to use the seed funding to improve its speech-to-speech models, expand its European training data, grow its sales and marketing teams, and scale infrastructure across European data centres.2 Those priorities align with its focus on European languages and on offering customers EU-hosted or self-hosted deployment options.