Such a product could soon be released: a wearable device that helps the wearer to manage anxiety and stress, aided by the latest in sophisticated artificial intelligence technology. The device will be able to collect a range of data relating to a person and then to offer alerts and advice. The device has been developed at the Massachusetts Institute of Technology.
The types of information collected includes speech patterns, analyzing the mood of a conversation, based on the speech patterns of the wearer, plus the tone of voice. The device also collects physiological information such as the heart rate, blood pressure, variations in skin temperature, or body movements. These signs and symptoms help to draw up a psychological and physiological profile; when these are compared with a database and interpreted by the computer housed inside the wearable, the device provides advice that aims to alleviate the stress.
Interviewed by Motherboard, one of the developers Mohammad Ghassemi explains more about the studies that led to the development of the device, focusing on the voice recognition and interpretation: “The way we set up our experiment was to explore if we had information on half of a conversation, could we reconstruct how they felt as they were telling these stories.”
The video below explains how the device will work in practice:
The studies involved recording physiological measurements and audio recordings from 31 volunteers; the participants also stated the ‘mood’ of what they were reciting for the audio recording, which often took the form of a story. The emotions were described on a scale of happy to sad. The stories were then broken down into five second excerpts, with each mini-segment being labeled as either happy, sad, or neutral. These segments were used to ‘teach’ the neural networks that form the core of the wearable, which formed the basis of the artificial intelligence. Overtime the neural networks learnt to associated tones of voice and key phrases with different human emotions and feelings. As an example, a ‘sad’ communication is more likely to be monotone and punctuated by elongated pauses.
For longer periods of communication, as outlined in a research paper, the device is able to guess the mood of a person with over 80 percent accuracy, although this drops considerably where the audio is short. To improve upon this the researchers aim to optimize the artificial intelligence by collecting more samples and further fine tuning.
Once the recognition aspect has been improved, the aim to market the wearable at people who suffer with anxiety or Asperger’s. Users can either replay conversations and receive feedback on how they came across, or receive ‘real-time’ signals if they begin to come across negatively while talking with another person (this is most likely to work when two people with a social anxiety condition are in conversation together).
