TÜBİTAK 1001: Development of a novel approach to hospital soundscapes incorporating an artificial intelligence (AI) model to predict users’ perception

Hospitals are one of the significant public spaces for receiving healthcare services. Therefore, hospitals should be designed to enhance the relaxation and well-being of patients, their relatives, and healthcare workers. To guide the design of a comfortable hospital environment, the user’s spatial perception must be correctly understood, and the emotional and perceptual clusters must be accurately captured. The proposed project aims to prepare a large dataset based on how users describe their hospital space/experiences/perceptions in their natural languages and to train an artificial intelligence (AI) model using natural language processing models. Parallel to this process, a Turkish database oriented towards hospital perception will be created, which will be one of the significant contributions of the project. This data will be used to retrain the model and obtain the components that form the perception. Similar data engineering will be done by collecting hospital interior photos and descriptions to complete the big data.

The hypothesis of the proposed project is as follows: while auditory perception in hospitals is a critical perception, there may be other factors/dimensions behind this perception depending on the user’s usage. Artificial intelligence models can reveal these underlying and missing perceptual dimensions. Tracing these factors and dimensions and identifying the missing/invisible parts can provide holistic clues for improving auditory perception and thus the hospital environment. It can answer the question of whether there is an unseen trace in the decision-support mechanism. This hypothesis will be investigated with four research questions: (1) Can the pre-prediction of the hospital auditory soundscape be made without visual-auditory stimulus data using natural language models? (2) How does the audio-visual environment in the hospital affect oncology clinic patients, their relatives, and healthcare personnel? (3) Is there any significant relationship between the psychoacoustic metrics of the acoustic environment in the hospital and the perceptual dimensions? (4) Can topics and perception clusters of participants be formed with the data collected from interviews with patients, their relatives, and healthcare personnel in the hospital?

As a method, two work packages will be conducted in parallel, and validations and gap fillings will be done with feedback data sharing. In WP1, text data in the literature and open-source datasets about the hospital soundscape will be analyzed using the BERTopic model based on the BERT (Bidirectional Encoder Representations from Transformers) language representation model developed by Google Language AI. It is aimed to establish a system where these data are combined with the visual data of the space using BERTopic and many other current models, ensuring the audio-visual integrity and relationships are present, trained, and the prediction model is created in the system. In WP2, face-to-face interviews/questionnaire surveys with hospital users (patients, their relatives, and healthcare personnel) will be conducted to determine what the hospital auditory perception is in real architectural space (example of Ankara City Hospital Oncology Clinic). Psychoacoustic metrics will be calculated by processing the sound recordings obtained from the oncology clinic. Qualitative and quantitative data correlations will be made with the qualitative state descriptions provided by the user’s subjective perception and the psychoacoustic metrics, which objectify the spatial perception. It is aimed to reveal/figure out the missing or background spatial perception by monitoring the data coming from the AI model in support of the qualitative-quantitative evaluations in the Oncology Clinic with tools measuring the hospital auditory soundscape.

One of the main outputs of the research project, the Turkish database, has the potential to grow as additions are made. The project output, which will work interactively, will feed the original AI model with the Turkish database, and as the number of architectural space cases increases and user interviews are decoded, our local database will develop further. This will be a very important database for our country as well. The proposed research is a study with great potential to create a audio-visual archive. Many researchers will be able to benefit from this database and audio-visual archive and prepare publications.