AUTHOR=Shibata Yuko , Victorino John Noel , Natsuyama Tomoya , Okamoto Naomichi , Yoshimura Reiji , Shibata Tomohiro TITLE=Estimation of subjective quality of life in schizophrenic patients using speech features JOURNAL=Frontiers in Rehabilitation Sciences VOLUME=Volume 4 - 2023 YEAR=2023 URL=https://www.frontiersin.org/journals/rehabilitation-sciences/articles/10.3389/fresc.2023.1121034 DOI=10.3389/fresc.2023.1121034 ISSN=2673-6861 ABSTRACT=In this study, we investigated a method to measure the subjective quality of life by using speech features for schizophrenia, which is the illness with the longest hospital stay in Japan. The issue in psychiatric treatment in Japan is the high rate of re-hospitalization and long-term hospitalization. Quality of life is an effective predictor of treatment but has not been utilized due to time constraints and lack of interest. In recent years, the development of voice-based applications has been actively developed. In this study, the Japanese Schizophrenia Quality of Life Scale (JSQLS) was measured using a conversational agent in 18 schizophrenic patients. The scores of the three subscales of the JSQLS, "Psychosocial," "Motivation and Energy," and "Symptoms and Side-effects," were used as objective variables, and the speech features (Chromagram, Mel spectrogram, Mel-Frequency Cepstrum Coefficient) during measurement was used as an explanatory variable. Comparing the accuracy of the ground truth and estimated values of the cross-sectional measurements, the mean value of RMSE was 14.518 and the k-nearest neighbors was the best. In the estimation of the scale scores of the six subjects measured longitudinally, the mean value of RMSE was 14.383, showing the best accuracy of Random Forest. In the cross-sectional measurement, nine out of 18 subjects, and in the longitudinal measurement, four out of six subjects were estimated with RMSE less than 10. Future studies using a larger number of data sets are needed to apply the results to telemedicine for continuous monitoring of QoL.