We have been listening to it for years: the development of artificial intelligence (AI) will produce autonomous robots that will be our slaves and, perhaps, will end up ousting us from our place as “kings of creation” if the fears of science fiction and some experts in robotics. But while that new era arrives, which does not seem as close as some proclaim, specialists are taking small steps that shorten the distance between machines and humans.
The New Scientist has echoed the work of Na Yang and Emre Eskimez, two graduates of the University of Rochester (New York) who have created software that not only captures the more or less hidden emotions in people’s words, but also that it does it better than humans themselves, something that has been achieved for the first time. Psychologist Melissa Sturge-Apple, also from the University of Rochester, intends to use this tool to better understand the emotions present in parent-child relationships and to unravel how they affect child development.
In the tests, the system created by Yang and Eskimez obtained a 72% success when classifying the emotional sense of 700 recordings, which surpasses the results of 138 workers of Amazon’s Mechanical Turk platform who were paid 50 cents. for every 10 samples they classified, a task in which they reaped 60% successes. The software also improved humans by identifying sound samples that were difficult to classify, a task in which it was 85% effective. This superior performance of the machine is not so strange, since in addition to hearing, people use other senses such as vision or even smell to unravel the feeling of others.
This new tool recognizes emotions as they arise, without having to record the speech . It identifies characteristics of the voice – tone, timbre, intensity, quantity, energy … – that its algorithms use to indicate the feelings behind them. These emotions – happiness, sadness, anger, fear, disgust or neutrality – are the only information that the software has in advance.
