Difference in speech analysis results by coding

Mental health disorder is becoming a social problem, and there is a need for technology that can easily check for states of stress and depression as a countermeasure. Conventional methods of diagnostic support and screening include self-administered psychological tests and use of biomarkers. However, there are problems such as burden on subjects, examination costs, dedicated reagents and equipment required for examinations, and reporting bias. On the other hand, voice-based evaluations are advantageous in terms of providing diagnostic support for physicians. They are non-invasive, do not require special and exclusive equipment, and can therefore be easily conducted remotely. We are pursuing the research and development of the Mind Monitoring System (MIMOSYS), which estimates the state of mental health from voice. Recorded audio is often compressed for efficient storage and transmission. However, there are concerns regarding the effects of deterioration of sound quality on analysis by MIMOSYS. Therefore, this study aims to verify the influence of the deterioration of voice quality due to coding on MIMOSYS analysis. As a verification experiment, coding was applied on the recording of 704 subjects reading 17 fixed phrases, assuming compression for transmission and storage. Then, the results of MIMOSYS analysis before and after encoding were compared. A strong correlation was observed before and after encoding, suggesting that MIMOSYS analysis is also valid for voice to which coding was applied.