A speech recognition application as a communication aid for acute and critical care patients with tracheostomies

Version 1 recorded patients mouthing predefined phrases. Version 2 introduced free-speech lipreading, converting silent speech directly into an audio playback. A prospective observational cohort study was conducted in Northern Ireland across three critical care units and one acute care ward.

Thirty-one adult patients were recruited between January 2023 and April 2024. SRAVI captured 468 video recordings, with Version 1 achieving 21.8% accuracy and Version 2 improving to 34.6%. Patients also used other communication methods, with unaided techniques (e.g., gestures and facial<br>expressions) comprising nearly half of all communication attempts.

Acceptability was explored through interviews with 35 healthcare professionals, nine patients, and five relatives. While attitudes toward SRAVI were generally positive, concerns were raised about patient acuity, Wi-Fi limitations, and device optimisation. Thematic analysis revealed key issues including emotional distress from ineffective communication and the need for holistic, patient-centred approaches.

In its current form, a larger trial of SRAVI is not feasible. However, iterative development and refinements may improve its clinical utility and support communication for this vulnerable population.