Fig 1.
A flow diagram of the article’s process including a) data collection and speech elicitation, which are then transformed into text transcriptions; b) text pre-processing and normalization, followed by data preparation for cross-validation in transfer learning to the model for each question to compute evaluation matrices; c) training of the model for LIME explanation and utilization of the weights from the explanation for creating word clouds for both the normal control group and the depression group.
Table 1.
The developed thai depression assessment tasks [43].
Table 2.
Participants’ demographic data.
Fig 2.
Samples of transcription of speech response from participants with original language (Thai) and English translation.
Table 3.
Recall, precision, sensitivity, accuracy, and numbers of support (positive predicted) for each group of six questions of reviews of symptom (Q1 − Q6) and six questions of describing emotional experiences.
Fig 3.
Word cloud of the responses to the question “How are you these days?” from six different groups: a) depressed females and males, b) depressed females, c) depressed males, d) normal control females and male, e) normal control females, and f) normal control males.