The ARTUS project delivered a fully integrated assistance system that automatically transcribes FM maritime radio communications and simultaneously identifies the transmitting vessel through radio direction finding. The system was successfully tested in a laboratory environment and demonstrators for both land‑based and ship‑mounted deployment are ready for installation. The core technical achievement lies in the combination of a speech‑recognition engine with a direction‑finding module, both of which were tightly coupled in a user‑friendly human‑machine interface designed with input from end users such as ship captains, dispatch staff, radio operators, and training instructors. Three application scenarios were defined to cover routine communication, emergency alerts, and mixed‑language exchanges, providing a comprehensive basis for system requirements.

Initially, the project used the Kaldi toolkit for automatic speech recognition. After a detailed evaluation of performance, the team switched to the Wav2Vec 2.0 model, which led to a noticeable reduction in word error rate (WER). Although exact numerical values are not disclosed in the report, the figures in the project documentation show a clear downward trend in WER following the technology change, indicating a substantial improvement in transcription accuracy. Data augmentation techniques, such as adding background noise, were applied to increase robustness, and a 4‑ears transcription principle was adopted to enhance the quality of the manually produced training data. The training set was split into dedicated training and test subsets, and the system’s performance was validated against reference transcripts.

Encryption and data security were integral to the workflow. All raw audio data were stored on an offline system at the German Rescue Radio (DGzRS) in a locked cabinet and encrypted with VeraCrypt. When the data were transferred to the subcontractor for training the speech recognizer, the transfer was also encrypted. The project leveraged Microsoft Teams, provided free of charge by Microsoft, for all remote meetings, workshops, and user‑testing sessions, which helped maintain collaboration during pandemic‑related restrictions.

The project’s timeline, originally planned for 2020–2022, experienced delays due to the COVID‑19 pandemic and the protective measures required by the rescue organizations involved. A four‑month extension was granted, during which the team integrated the speech recognizer into the early demonstrator, hired an additional research assistant for transcription, and established a simulation environment for testing. Despite these adjustments, all milestone objectives were met, and the overall budget remained within the allocated plan, with savings in demonstrator production, travel, and software licensing offsetting the increased personnel cost.

Collaboration was structured around three main partners: the German Rescue Radio (DGzRS), the Institute for Applied Information Systems (IAIS) as subcontractor, and a consortium of additional research institutions. DGzRS participated in all four sub‑work packages, providing operational expertise and field testing sites. IAIS supplied the speech‑recognition technology and handled the data‑processing pipeline. The consortium contributed to system design, scenario definition, and user‑interface development. The project was funded under a German federal research program, with the total expenditure staying below the planned budget except for the additional personnel cost, which was compensated by the aforementioned savings.

After the laboratory validation, the next phase will involve field trials in real operational contexts. The partners have agreed to continue collaboration beyond the official project end, particularly to further develop and refine the speech‑recognition component. The ARTUS system therefore represents a significant step toward automated situational awareness for maritime search and rescue operations, combining robust transcription, accurate source identification, and a user‑centric interface within a secure and cost‑effective framework.