"Voice AI": Corti and Dedalus cooperate for documentation in clinics

The Danish health tech company Corti and the software provider Dedalus have announced a partnership for medical documentation in hospitals using voice input and AI. Corti's AI infrastructure is to be integrated into the new Orbis Speech Assistant from Dedalus and could serve as an alternative to existing solutions from major US providers such as Microsoft.

In the future, the clinical software will relieve the burden on medical staff by analyzing spoken content from patient conversations in real time, automatically structuring it and providing it as clinically relevant documentation. The hope is to significantly reduce administrative work and increase the quality of documentation. The plan is to avoid typical sources of error such as inaccurate or overly extensive texts, which often still occur with generic AI systems. By integrating the AI model into the Orbis Speech Assistant, the company aims to "set a new standard for clinical documentation", according to Jan Rusch, Head of Integrated Technologies at Dedalus.

There are now numerous medical software providers that process doctor-patient conversations using AI speech recognition. Since January 2025, the University Medical Center Hamburg-Eppendorf (UKE) has also been using the AI speech recognition system Aureon, which was developed by UKE subsidiary IDM gGmbH. Aureon will also be available as a cloud solution in other hospitals in the future. "In our opinion, data collected in the German healthcare system should primarily also help our patients and colleagues here in Germany to provide even better, safer and more efficient medicine. Models developed in Germany make the German healthcare system less dependent on non-European providers," says IDM Managing Director Dr. Julius Obergassel.

The start-up "Voize" has also developed artificial intelligence for speech recognition and documentation, which structures the input and automatically creates the appropriate documentation entries. The services are also increasingly being used by nursing staff. Care staff can use an app to enter their documentation by voice. The AI processes voice input locally on the smartphone.

LLMs have limits

ID Berlin, for example, also offers services for natural language processing – whether in writing or speech as input. The focus there is on the automated processing of texts to make billing processes more efficient. Both rule-based systems and machine learning (ML) approaches have been used for decades, as Dr. André Sander explained to heise online when asked. "The availability of LLMs has significantly improved certain aspects of text processing, but certain elements of the processing chain (for example medical, comprehensible plausibility checks or transparent reasoning) are still mapped using ontologies. These are by no means relics from the 'pre-LLM era', but ultra-modern structures that are being further developed by an active community worldwide", says Sander. He also refers to the limitations of LLMs and a study recently published by Apple's ML research group, according to which the "thinking" of large reasoning models could be at least partly an illusion.

This fits in with the wishes expressed at the German Medical Conference. Among other things, they emphasized the importance of digital sovereignty in the healthcare sector. They wanted to be able to develop and train European AI models so that sensitive medical data is not passed on to large technology companies, which in turn use this data to develop or improve their products.

(mack)