€7.7m · Seed · Software · Berlin, Germany
Berlin-based Deepslate has raised €7.7m in seed funding led by 42CAP, with participation from Alstin Capital, existing investor SIVentures and several business angels. The company will use the capital to expand its European training-data programme, improve latency and speech quality, build sales and marketing, and scale production infrastructure across European data centres, according to Tech.eu.
A direct speech model removes the middle layers
Most voice agents assemble three systems: speech recognition converts audio to text, a language model produces an answer, and text-to-speech turns that answer back into audio. Deepslate instead trains a direct speech-to-speech model. The company says this preserves cues such as emphasis and intonation while reducing the delay introduced by moving through separate models.
Its architecture still has distinct components: a speech encoder, a reasoning core based on an open-weights language model, and a speech decoder. The commercial idea is that Deepslate can improve or swap the reasoning layer without retraining the entire audio system. That could shorten model-upgrade cycles while allowing the company to specialise the speech layers for European names, addresses, accents and dialects.
The product is delivered through a self-service platform and developer integrations including API, SIP, WebRTC, LiveKit and Pipecat. Deepslate's site lists 27 supported languages and says its model recorded 440 milliseconds of latency in an Artificial Analysis comparison. Those performance figures remain company claims reported by Tech.eu, but they point to the product's core sales argument: making automated conversations responsive enough for live customer interactions.
Deployment control is part of the product
Deepslate says insurers, contact centres and software platforms already use the technology in production, although it has not disclosed customer names or revenue. For larger deployments, customers can use European cloud infrastructure or run the model in their own environment. The company presents self-hosting and EU data residency as a route into regulated or data-sensitive workloads, rather than as a compliance layer added after the model is built.
That creates a clear execution test for the seed round. Deepslate now has to convert benchmark speed and European-language performance into repeatable enterprise deployments while supporting multiple hosting models. Expanding training data and infrastructure should improve the technical product; growing sales capacity will show whether those differences are strong enough to win and retain production workloads.
Sources checked:
Tech.eu · Deepslate · Deepslate product · Deepslate security · StartupValley · Tech.eu Funding Explorer
