A homogeneous information system facilitates cooperation between departments and the management of the entire infrastructure. The voice recognition system must be integrable into different management systems (RIS, HIS, PACS, EPR, etc.) to optimize existing workflows without disrupting work habits.

The accuracy of speech recognition depends largely on the vocabulary and the language model. A good system must provide maximum robustness in generic fields and be able to be optimized for individual disciplines. The multidisciplinary medical vocabulary released for Italy currently covers more than 90% of the medical terminology used in the entire hospital structure and can be further extended with customizations and onsite pre-adaptations. In the recomed architecture, the vocabularies are shared so that the automatic adaptation of each doctor can contribute to increasing performance for everyone.

Stability of a system, in a hospital environment, is essential. It means that the systems, data, network connection, etc. are all operational, stable and secure, to be able to carry out activities without interruptions. Add to this the ability of the software to remain functional and consistent over time without undergoing degradation. A software, developed with FMEA methodology, like that of reconice, guarantees exceptional stability, by design.

A distributed network solution must allow centralized management of the entire system, ensuring maximum flexibility for administrators. Management and configuration tools must be accessible from any workstation. Each physician must be able to access his or her voice profile, settings, presets, etc. from anywhere to report from any workstation.

Each department has its own needs. The use of mobile systems (notebooks, tablets, etc.) is increasing even in the healthcare sector. A modern voice recognition system must be able to be used on all necessary devices, in all modes (live, streaming or deferred), regardless of the client's operating system, guaranteeing all the features and the best performance in any context. It must also allow its use with any acquisition device, from traditional microphones with control buttons to headphones with or without wires, for those who need to keep their hands free.

The first thing is accuracy, which depends on the entire chain, from signal acquisition (microphone) to the engine, through the Language Resources. Secondly, the real-time factor is important, that is, the time lapse between dictation and the transcribed text. Thirdly, the features to optimize the flow, from macros to correction tools.

A speech recognition system is a long-term investment and therefore must be able to grow within the organization. It must be scalable in terms of size, vocabularies and also functionality. It must work on any device, with any OS, on geographic networks, on a large scale (even for thousands of users) on physical or virtual servers, with load balancing and fault tolerance mechanisms.

Among the advantages of adopting a voice recognition system, the possibility of optimizing the workflow stands out. A good reporting system must OPTIMIZE and never distort the workflow. It must adapt to the way doctors work and must not be a Business Process Reengineering tool, imposing procedures. It must also be equipped with adequate verification and correction tools, so as to be able to dictate, correct and release the document in a short time and with the highest quality.

SECURITY is fundamental. "Security by design" is an essential requirement: the measures to be adopted to guarantee the confidentiality, integrity and availability of software systems during the entire life cycle of software development cannot be an afterthought. The certification of the supplier company for quality standards throughout the entire production, sales and maintenance cycle, guarantee a constant risk analysis, the best coding practices and procedures that promptly identify threats, rigorously testing. And of course the penetration tests, entrusted to industry leaders, for periodic, constant checks. Reconice is a ScienceSoft partner for this activity.