Search

Search Results

• EMOTION DETECTION AND CLASSIFICATION SYSTEM BASED ON SOUND FLOW DATA

  А.А. Egorchev, D. М. Pashin, N. А. Sarambaev, А. F. Fakhrutdinov

  2024-10-08

  Abstract ▼

  In today's rapidly changing and demanding work environment, the ability to quickly and accurately
  assess an employee's emotional state is crucial to protecting human lives and reducing material risks.
  Emotional well-being plays an important role in workplace safety, productivity, and overall mental health.
  Therefore, the development of effective tools for monitoring negative emotions and responding to them is
  an urgent task of our time. The purpose of this study is to develop an algorithm capable of classifying
  emotions using audio data recorded by a user's smartphone. Such a tool is especially useful if integrated
  into a broader health monitoring system that allows you to evaluate human health indicators in real time
  using non-invasive methods. This article presents a new solution that uses acoustic signals picked up by a
  smartphone microphone to detect and classify user emotions. Using convolutional neural networks
  (CNNS), a type of deep learning algorithm known for its effectiveness in processing audio and visual data,
  the proposed system can determine the user's emotional state. The CNN model is trained to recognize
  patterns in audio data corresponding to various emotional manifestations, focusing on detecting negative
  emotions such as anger or sadness. The results of the study demonstrate the effectiveness of the system:
  the error rate in determining negative emotions is 19.5% for false positive results (errors of the first kind)
  and 20.1% for false negative results (errors of the second kind). These indicators indicate its potential for
  practical application in real conditions. By integrating this solution into existing biomedical monitoring
  systems, organizations can expand their ability to monitor the emotional well-being of employees, potentially
  preventing negative consequences such as industrial accidents or mental health crises. The integration
  of emotion recognition using smartphones into health monitoring systems represents significant progress
  in the field of non-invasive biomedical monitoring, using the ubiquitous presence of smartphones
  and machine learning capabilities.

• DEVELOPMENT OF A SYSTEM FOR DETERMINING THE FREQUENCY OF SURFACE RESPIRATION ACCORDING TO SMARTPHONE BIOMETRIC SENSORS

  D.Е. Chikrin, A.A. Egorchev, D. М. Pashin, N.А. Sarambaev

  2023-10-23

  Abstract ▼

  Modern realities set digitalization tasks for humanity in various areas of work and life,
  speeding up the deadline for completing tasks and facilitating them. Modern technology equipped
  with sensors that can be used for preliminary diagnosis allows you to identify various symptoms
  that may be the reason for visiting medical institutions. This allows you to gain time – a very valuable
  resource when it comes to human life. Therefore, the ability to perform such diagnostics the
  determination of respiratory rate, is an urgent task today. The article presents a method for determining
  the respiratory rate using a three-axis accelerometer on a mobile device. This method can
  be used in a user health monitoring application in the absence of smart watches. The method allows
  the user to measure the respiratory rate of a person only if the user is in a sitting position and
  a mobile device equipped with the necessary sensor is in the upper anterior thigh area (pocket
  area). The algorithm for determining the respiratory rate is implemented in two programming
  languages: Python and MatLab. The algorithm uses a respiratory rate stabilizer, because the accelerometer sampling rate is not constant from an Android-based mobile device. Next, the signal is
  normalized by the z-normalization method. To isolate the frequency interval in which the respiratory
  rate is calculated, the Butterworth filter of the 1st order is used. The analysis of independent
  components makes it possible to obtain its independent components from a mixture of signals.
  Several implementations of this method have been tested in Python and Matlab. The best quality
  results were shown by an algorithm implemented in MatLab using the built-in reconstructive analysis
  of independent components (RICA) from a set of statistics and machine learning tools. In
  terms of speed, the best results were shown by the implementation of the algorithm in Python with
  the method of fast analysis of independent components (FastICA). The MSE for the range of
  10-20 breaths per minute was 2.14 breaths per minute. The MSE for 20-30 breaths per minute was
  3.46 breaths per minute.