Research On Speech Segmentation Algorithm Based On Long-Term Wearable Social Data Set

With increase of stress in work and study environments,mental health issue has become a major subject in current social interaction research.Generally,researchers analyze psychological health states by using the social perception behavior.Speech signal processing is an important research direction as it can objectively assess the mental health of a person from social sensing through the extraction and analysis of speech features.In this paper,a series of four-week long-term social monitoring experiment study using the proposed wearable device has been conducted.A set of Wellbeing questionnaires among of a group of students is employed to objectively generate a relationship between physical and mental health with segmented speech-social features in completely natural daily situation.In particular,we have developed transfer learning for acoustic classification.By training the model on TUT Acoustic Scenes 2017 dataset,the model learns the basic scene features.Through transfer learning,the model is transferred to the audio segmentation process using only four wearable speech-social features(Energy,Entropy,Brightness,Formant).The obtained results have shown promising results in classifying various acoustic scenes in unconstrained and natural situations using the wearable longterm speech-social dataset.1)Designing a wearable intelligent device to evaluate mental health based on long term speech-social feature analysis which can deal with speech signals in persuasive manner.In order to unobtrusively and continuously marking the granular details of behaviors and contexts,limited four speech-social feature signals are used for analysis and embedded to avoid directly recording their voice.We designed an experiment for university students to collect data for a long period of one month for evaluating their physical and mental health.2)The parameter/model transfer learning is applied in the field of acoustic classification and its effect is studied through a model-based transfer learning algorithm.The transfer learning model is verified to solve the problem of insufficient training samples as it often happens in the traditional model.3)Finally,the interrelationships between the segmented speech-social features with anxiety level state are established.The results are used to assist in studying the mental health of college students.