Designed by the Russian company Neurodata Lab from scratch prepared as a database for neural network learning, and technology to assess the emotions of people on the basis of the set parameters according to “Izvestia”. The first stage was shot 6 hours of video, in which 10 professional actors (they are in the process recorded the breathing pulse) depicted on the record in a neutral mood, joy, surprise, anger, sadness, fear and disgust. Then, the resulting footage was studied annotators who reported observed in specific moments of emotion.
The next step was the selection of 160 hours of video, which depicted communicating in a different environment people. With the help of thousands of annotators from all over the world were able to identify 20 emotions. Specifically for this task developed a platform Emotion Miner who continues to work and allows you to analyze video.
The collected data formed the basis of techniques for training neural networks to recognize human emotions. The algorithms pay attention to the voice (pitch, tone, volume, pauses in speech), emotional and semantics of the text, facial expressions, speed and direction of body movements and positions of the individual limbs, the heart rate based on changes of skin color, respiration motion of the chest, as well as gender, age of person and the presence of on her face glasses, moustaches, beards.
“We have combined convolutional and recurrent neural networks and the result was a multimodal architecture that could simultaneously analyze audio, video, body movements and physiological parameters. The accuracy of recognition of emotion with neural network greatly increased”, — told “Izvestia” technical Director Neurodata Lab Andrey Belyaev.
The development, which is already ready to use, can help in the fields of business, advertising, security and medicine, as well as other activities where it is important to understand the human behaviour and emotions.
The cure for schizophrenia increased bust of up to 3-gauge