Fig. 1

Outline of the study procedure. Acoustic features were extracted using pyAudioAnalysis and DisVoice audio python libraries. Audios were transcribed using Google Speech-to-Text API. Linguistic features were extracted using LIWC. POS and word frequency features were extracted using NLTK. Sentiment and tone analysis was performed using NLTK, Watson Tone Analyzer, Azure Text Analytics, and Google NLP. We perform an ensemble feature selection to identify a subset of predictive features. We use different machine learning and deep learning techniques to build a suicidal ideation classification model