Deepslate secures €7.7M for its European voice AI platform

Berlin-based voice AI firm Deepslate has successfully raised €7.7 million in seed funding, with the majority of the investment coming from Munich's 42CAP. The round also saw backing from Alstin Capital, existing investor SIVentures, several business angels, and individual investors. Deepslate specializes in developing its own speech-to-speech AI models, which work directly on spoken language rather than first converting it into text.

This approach is particularly focused on European languages and ensures that the technology remains within the EU.

Conventional voice AI typically translates speech into text, processes it through a language model, and then turns the response back into speech. In contrast, Deepslate's end-to-end model allows audio to be processed directly and returned as audio. This method reduces latency while maintaining the nuances of spoken language, such as tone, emphasis, dialect, and other aspects.

The company's architecture comprises three components: a speech encoder, a reasoning core based on an open-weights language model that Deepslate fine-tunes for specific languages, and a speech decoder that produces the spoken response. This architecture enables the underlying language model to be swapped out without the need for retraining the entire system.

Deepslate's model reportedly achieved a response time of 440 milliseconds in an independent benchmark, making it the fastest speech-to-speech model measured. Additionally, the technology achieved the best error rate in the comparison field for European languages in the CoVoST2 benchmark. Deepslate is ISO 27001 certified, ensuring robust data security practices.

The company's solution is already in use by insurers, contact centres, and platforms in production. Deepslate offers the model via a self-service platform and API, with volume and self-hosting options available for larger enterprise customers who prioritize data sovereignty. Deepslate's co-founder, Paskal Paesler, emphasized that data sovereignty is not just a preference but a prerequisite for their customers.

To address this, Deepslate provides transparency about where the computing occurs, the identity of its subprocessors, and details on open-source components.

With the new funding, Deepslate plans to expand its model training and European data initiatives, with a particular emphasis on German street names, personal names, and dialects. The company will also focus on reducing latency and enhancing voice quality. Additionally, Deepslate will bolster its sales and marketing teams and scale its production infrastructure across European data centres to accommodate increasing call volumes.