Deepslate develops its own AI voice models for real-time speech processing

Deepslate develops its own AI models that generate voices in real time, competing with OpenAI and Elevenlabs. The start-up has now raised millions for its approach.

Deepslate founder Paskal Paesler (l.) and Jan Brachthäuer: “High demand” for their models. Photo: Company

Dusseldorf. Voice assistants are still causing disillusionment among many customers. They misunderstand concerns, react slowly or lead callers through rigid dialogues. Particularly on the telephone, fractions of a second count. Nevertheless, the market for AI-generated voices, so-called Voice AI, is growing rapidly.

Investors around the world are betting that it will be easier for people to express their thoughts rather than first writing them down and typing them into a chat window.

However, it is more demanding for AI systems to process spoken language: Speech-to-speech models have to understand the language, generate a response and convert it back into speech - and if possible without any noticeable delay. This is exactly the problem that the Berlin start-up Deepslate wants to solve.

The start-up develops models that process spoken language in real time and output it as language. The founders Paskal Paesler and Jan Brachthäuser are competing against financially strong competitors such as OpenAI and Elevenlabs. Deepslate therefore uses a different training approach.

Voice AI market is growing

Since the beginning of the year, venture capitalists worldwide have invested more than $20 billion in companies developing AI-generated speech solutions.

More on the topic of our partners display

remind.me Take advantage of current low electricity/gas prices before prices rise again