How AI can improve PSAP operator training for Emergency Call Handling

Training operators within Emergency Control Rooms plays a critical role in ensuring optimal call handling. It requires specialized and targeted preparation tailored to the different types of emergency services (112, 118, 115, 116 117, etc.). However, training is not limited to domain-specific knowledge alone; it also encompasses behavioral and psychological aspects.

Emotional management is one of the most important of these. Callers seeking assistance are, in most cases, experiencing high emotional distress, which significantly affects communication quality. At the same time, operators themselves may be exposed to stress and crisis situations, particularly during call surges such as large-scale emergencies. This takes place in a broader context of increasing pressure on Emergency Control Rooms, driven by rising call volumes, growing citizen expectations, and a shortage of specialized personnel, as highlighted by research conducted by EENA – the European Emergency Number Association. Operator training is therefore more critical than ever to ensure an adequate and reliable service to the community.

Operator training: the contribution of Artificial Intelligence

In recent years, several pilot projects and experiments have explored the use of Artificial Intelligence within Emergency Control Rooms, as documented by EENA. These initiatives have helped identify the areas where AI implementation is most effective. As a result, significant contributions have emerged in the field of operator training, particularly through the use of Virtual Experts and scenario simulation.

Virtual Expert

A Virtual Expert is an AI model trained on a specific knowledge domain. It can therefore provide real-time responses to concrete operational situations, such as suggesting which questions to ask a caller or analyzing their responses. In the context of operator training, a Virtual Expert can be used to deliver continuously updated content related to the use of technological tools, operational procedures, regulatory frameworks, and legal references.

Scenario simulation

The level of preparedness of Emergency Control Room operators is directly proportional to their field experience. AI, leveraging neural architectures such as Transformers (networks designed to efficiently process data sequences like text or audio), makes it possible to recreate realistic emergency scenarios. These simulations are generated by AI models trained on historical real-world cases, with scenarios appropriately adapted and enriched to support training objectives.

AI as support for Call-Taker interviewing

Conducting an effective interview with the caller is one of the pillars of emergency management. During this phase, operators face several challenges, including caller distress, limited communication skills, language barriers (local dialects, foreign languages, or mixed usage), and environmental noise.

In these areas, Artificial Intelligence should be applied selectively:

• Speech-to-Text: AI algorithms provide valuable support for transcription and keyword analysis. However, error rates caused by disorganized or emotional speech must always be taken into account.

• Emotion Management: Voice tone and breathing patterns (e.g., shortness of breath) can be analyzed using specialized models such as Voice Emotion Recognition (VER).

• Background Noise Reduction: AI-based speech enhancement models can identify vocal patterns and effectively filter non-stationary noise such as traffic, wind, or people shouting.

It should also be noted that these models can improve their reliability over time through adaptive mechanisms (voice characteristics, linguistic style, emotional expression), allowing them to personalize their behavior to the individual caller.

How to implement a Virtual Expert in 4 steps

Implementing a Virtual Expert within an Emergency Control Room is highly dependent on the operational context, which varies by emergency type and professional roles involved. A typical use case is a call related to a suspected cardiac arrest. A Virtual Expert should be implemented through a structured four-step process:

1. LLM Training. Starting from a base LLM, the AI acquires deep domain expertise through the analysis of official documents and selected resources: regulations, protocols, guidelines, clinical cases, reference manuals, and more.

2. Operational context acquisition. The Virtual Expert operates within a real-world environment composed of data flows, operator roles, interfaces with existing software, and decision-support dashboards. This information provides the system with situational awareness.

3. Reasoning implementation. Different logic mechanisms (rule engines, inference systems, etc.) are used to interpret the information produced by the LLM and to support decisions based on predefined guidelines, or to recommend specific actions (e.g., triggering a high-priority response).

4. Ethical framework adoption. The use of AI in emergency settings requires clear ethical rules to ensure safe behavior and build operator trust. Humans must remain the final decision-makers (Human in the Loop) to prevent cognitive bias and ensure accountability.

The use of AI in Emergency Control Rooms is no longer optional but inevitable, given its vast potential to support operations. In the Emergency Control Room of the future, close collaboration between AI systems and human operators will be essential—provided there is continuous validation of model performance and regular updates to ensure AI outputs remain accurate, relevant, and resilient to change.