• Here is my summarization of sociometer research paper.
  • This paper specifically tell you on how to extract voice data for sociometric calculation and on how to detect if face to face conversation happens.
  • So, this is great paper to follow if you want to go deep into the face to face detection and speech recognition.
  • Below is the .pdf.

./20161112-0959-cet-state-of-the-art-13-sociometer-1.pdf

./20161112-0959-cet-state-of-the-art-13-sociometer-2.png

./20161112-0959-cet-state-of-the-art-13-sociometer-3.png

  • Sociometer is a wearable sensor package for measuring face to face interactions between people.

./20161112-0959-cet-state-of-the-art-13-sociometer-4.png

  • Sociometric is important in these fields.
    • Knowledge management application for finding expert.
    • Organizational behavior.
    • Social network analysis.

./20161112-0959-cet-state-of-the-art-13-sociometer-5.png

  • Especially noted that I also had a similar idea about this people Wikipedia, where you can search informations based on the person.

./20161112-0959-cet-state-of-the-art-13-sociometer-6.png

  • The sociometer and the Sociometric Badge are exist to minimize the usages of these things.
    • Diaries (what?!?!?!).
    • Questionnaires.
    • Surveys.

./20161112-0959-cet-state-of-the-art-13-sociometer-7.png

./20161112-0959-cet-state-of-the-art-13-sociometer-8.png

  • Nature of communication is important to understand the following phenomenas.
    • Diffusion on informations.
    • Group problem solving.
    • Consensus building.
    • Coalition information.
  • Neither of these has a place in my memory.

./20161112-0959-cet-state-of-the-art-13-sociometer-9.png

  • Knowing physical structure of an institution can either encourage or hinder productivity.

./20161112-0959-cet-state-of-the-art-13-sociometer-10.png

  • Wearable sensors with pattern recognition will play important role in modeling and sensing physical interactions.

./20161112-0959-cet-state-of-the-art-13-sociometer-11.png

  • There were no other option to gather interactions data at this moment this paper is written.
  • This is because it is hard to obtain reliable data out of social activity.

./20161112-0959-cet-state-of-the-art-13-sociometer-12.png

  • They used machine learning to understand the pattern.
  • I believe this is the oldest thing I see term "machine learning" mentioned.
  • I guess I need start to look back to see what kind of things will popular in the future.

./20161112-0959-cet-state-of-the-art-13-sociometer-13.png

  • Interaction state like these.
    • The duration of the conversation.
    • To whom this person are talking to.
  • These informations above solely can be used to infer dynamic and social structures.
  • This method should be cheaper and efficient than human driven interview or questionnaire.

./20161112-0959-cet-state-of-the-art-13-sociometer-14.png

  • Also sociometer and its iteration are hoped to make social research to be easily scalable to larger group.

./20161112-0959-cet-state-of-the-art-13-sociometer-15.png

  • In summary they are looking to discover on how information about social network relationship can be derived by statistical machine learning.

./20161112-0959-cet-state-of-the-art-13-sociometer-16.png

  • To know identity of to whom this person is talking, sociometer uses IR transceiver to broadcast information.
  • This is good to mention actually because in my current state of my bachelor thesis my badge has no idea to whom this people are talking to.

./20161112-0959-cet-state-of-the-art-13-sociometer-17.png

  • The sociometer was designed by wearable designer Brian Clarkson (Google search returns nothing related, or I missed things).
  • Although the sociometer was designed to have these in mind.
    • Aesthetic.
    • Comfort.
    • Optimal sensors placement.
  • However, it did not prevent its wearers to stop using it because it is not "wearable" enough.

./20161112-0959-cet-state-of-the-art-13-sociometer-18.png

  • Some specifications of the sociometer.
    • 2 accelerometers.
    • 256 MB on board storage.
    • IR transceiver.
    • Microphone.
    • Power supply.
    • Powered by 4 AAA batteries.
    • Shoulder mount.

./20161112-0959-cet-state-of-the-art-13-sociometer-19.png

  • Basically this device store information about these things.
    • Information about nearby people.
    • Motion information.
    • Speech information.

./20161112-0959-cet-state-of-the-art-13-sociometer-20.png

  • Other sensors can be added as well. For example like these.
    • Light sensor.
    • GPS.
  • It is mentioned here that the researchers are not yet to use the data from motion sensors.

./20161112-0959-cet-state-of-the-art-13-sociometer-21.png

  • Sociometer hardware.

./20161112-0959-cet-state-of-the-art-13-sociometer-22.png

  • The use of low powered IR transceiver is optimal because of these matters.
    • Only detect people that is facing the sociometer wearer.
    • Sociometer only detects people in close range.

./20161112-0959-cet-state-of-the-art-13-sociometer-23.png

  • The transceiver create a cone shaped region in front of its wearer.

./20161112-0959-cet-state-of-the-art-13-sociometer-24.png

  • IR range detection is approximately 6 feet.

./20161112-0959-cet-state-of-the-art-13-sociometer-25.png

  • Here are some concern about privacy.
  • However, the researchers mentioned here that the data taken is only the important parameter without the context.
  • For example for sound, the data extracted is only the features (pitch, volume, ...) without the contextual speech itself.

./20161112-0959-cet-state-of-the-art-13-sociometer-26.png

  • It is called to garbling the audio.
  • To make the audio features are all the same.
  • But "encrypt" the voice.

./20161112-0959-cet-state-of-the-art-13-sociometer-27.png

  • There was a problem with IR transceiver because the signals are not consistent.
  • This is due to nature of human conversation that is not always face to face.

./20161112-0959-cet-state-of-the-art-13-sociometer-28.png

  • There was also a problem that the IR transceiver detects face to face communication with its own wearer.

./20161112-0959-cet-state-of-the-art-13-sociometer-29.png

  • Above is a quite specific explanation on how to calculate face to face detection.
  • I do not read these things yet.

./20161112-0959-cet-state-of-the-art-13-sociometer-30.png

  • Above is a quite specific explanation on how to combine face to face detector with the speech detector.
  • These are all to detect if a conversation happen.
  • Whether or not the sociometer wearer just blabbering around or having an actual conversation.
  • I do not read these things fully yet, nor I understand these.

./20161112-0959-cet-state-of-the-art-13-sociometer-31.png

  • Simple summation on the social noise.

./20161112-0959-cet-state-of-the-art-13-sociometer-32.png

  • Diagrams.
  • I am not sure what are these for.
  • But, these display on simplest sociometric flow. Please check into Moreno's Sociogram.