SAFERS: Speech Analytics for Emergency Response Services

Hjem/Prosjekter/Pågående prosjekter/SAFERS: Speech Analytics for Emergency Response Services

Project title: SAFERS: Speech Analytics for Emergency Response Services

Primary and secondary objectives of the project

The main objective of the SAFERS project is to develop a set of innovative technological solutions enabling the real-time analysis of phone calls to emergency response services. This objective includes (1) the automatic transcription of the calls using speech recognition, (2) the extraction  of high-level semantic information from these dialogues, and (3) the prediction of critical variables about the emergency situation based on contextual factors. The project will implement a software prototype that can be used to assist the emergency operators with real-time analysis, redictions, and automatic error detection.

The SAFERS project also aims to advance the state-of-the-art in research areas such as speech recognition, semantic parsing, dialogue modelling and machine learning. A final objective of the project is to foster new, tight collaborations between researchers in these areas and the  healthcare sector at large, thereby facilitating the transfer of technological expertise.

Project summary

The SAFERS project will develop a set of innovative tools and techniques for performing real-time analysis of emergency calls. These tools and techniques will build on recent technological advances in speech processing, language technology and machine learning, with the goal of enhancing the quality, efficiency and safety of emergency responses.

The project will need to address several important R&D challenges. One major research effort will be the development of a speech recognition system for Norwegian able to transcribe emergency calls with sufficient accuracy. The project will notably investigate how to integrate environmental noise, psychological stress and other emotional factors into the acoustic models of the speech recogniser. Dialogue modelling techniques will be applied to track in real-time the interactions between the caller and the emergency operator and automatically extract important pieces of (medical) information from the dialogues. Finally, the project will apply machine learning techniques to automatically predict relevant variables (such as the situation’s urgency level or the most likely geographical position of the caller) on the basis of the information collected during the call, combined with various contextual factors. These predictions will be subsequently exploited to quickly detect potential errors, omissions or deviations from operational guidelines.

The SAFERS project brings together an international consortium of leading researchers in the fields of speech recognition, natural language processing, statistical modelling and machine learning. The project’s stakeholders (represented by the National Centre on Emergency Communication in Health) will also take an active role in the SAFERS project, both regarding the design of technological solutions that are best suited to the practical needs of emergency response services, and the empirical evaluation of
these solutions in real-life scenarios.

Popular science presentation

Emergency response services – such as 113 for medical emergencies in Norway – must operate in difficult and time-critical conditions. Medical emergencies are characterised by multiple uncertainties, as the operators need to assess a complex medical situation based on a single information source, namely the patient, relative or caregiver on the phone. The operators must then quickly decide how to handle the situation and dispatch medical resources (such as doctors or ambulances) accordingly.

The goal of the SAFERS project is to apply state-of-the art techniques within speech processing, language technology and machine learning to help make emergency response services more reliable and effective. The project seeks to develop a system able to process and analyse in real-time the spoken interactions between the emergency operators and the callers in order to:
(1) Identify what is being said (important keywords and factual answers) during the course of the interaction;
(2) Transcribe in real-time the content of the conversation using speech recognition;
(3) Exploit this information together with contextual factors to predict important parameters, such as the urgency level of the situation;
(4) Detect potential errors or omissions in the emergency response.

The technological solutions developed by the SAFERS project hope to assist emergency operators with real-time analysis and predictions, allowing them to focus on assessing the medical situation and providing assistance to the callers. These solutions will also provide an additional layer of control to quickly detect potential anomalies or important deviations from operational guidelines. Finally, the transcription tools will enable emergency communication centres to simplify and streamline the documentation process.


Prosjekt-tittel: SAFERS: Talegjenkjenning og maskinlæring for nødmeldetjenester

Norsk oversettelse:

Nødtjenester, som 113 for øyeblikkelig helsehjelp i Norge, må fungere i vanskelige og tidskritiske situasjoner. Ved en medisinsk nødsituasjon kan det være mange usikre faktorer, og operatøren er ofte nødt til å vurdere en vanskelig og kompleks medisinsk situasjon basert på kun en informasjonskilde, nemlig innringeren, som kan være pasienten selv, en slektning eller annen omsorgsperson. Operatøren må på bakgrunn av denne informasjonen gjøre beslutninger for hvordan situasjonen best kan håndteres, og eventuelt sende ut medisinske ressurser som lege eller ambulanse.

SAFERS-prosjektet har som hovedmål å benytte moderne teknologi innen taleprosessering, språkteknologi og maskinlæring til å forbedre nødtjenestene slik at påliteligheten og effektiviteten kan økes. Mer presist så ønsker prosjektet å utvikle teknologi som kan behandle og analysere nødtjenestesamtaler mellom medisinsk personell og innrigere i sanntid. Et slikt system skal:
(1) identifisere hva som blir sagt (nøkkelord og viktige fakta) underveis i samtalen.
(2) transkribere samtalen i sanntid ved hjelp av talegjenkjenningsteknologi
(3) utnytte informasjon fra samtalen, sammen med kontekstfaktorer, til å bestemme viktige parametere, for eksempel hvor tidskritisk situasjonen er.
(4) Oppdage potensielle feil og mangler i behandlingen og vurderingene av nødanropet.

De teknologiske nyvinningene fra SAFERS-prosjektet vil bistå nødtjenesteoperatorer med sanntidsanalyser og prediksjoner, slikt at de kan ha fullt fokus på medisinske vurderinger og gi riktig assistanse til innringere. Disse nyvinningene vil også tilby et ekstra kontrollnivå ved at anomalier og større avvik fra operasjonelle retningslinjer kan oppdages raskt. Transkriberingsverktøy vil også kunne føre til en enklere og mer strømlinjeformet dokumenteringsprosess for AMK-sentraler.