Algorithms detect tone of voice, helping to determine improvement or deterioration in interactions
Every personal relationship is bound to hit a rough patch, sooner or later. Some recover, while others disintegrate. To stop this from happening, therapists now have a new tool at their disposal: artificial intelligence, or AI. The framework designed by researchers from the University of Southern California and the University of Utah can estimate the length of each individual relationship by analyzing the tone of voice that partners use when addressing each other during problem-solving interactions. These acoustic features, once fed into the algorithms, provide cues for determining if there’s been any improvement (or deterioration) since the last therapy appointment, as well as the change in the relationship status.
To provide data for machine learning, scientists gathered information from 134 couples, a total of 229 recorded audio samples. Features such as pitch, intensity, jitter and shimmer were used as essential speech properties and primary parameters for audio analysis. (Jitter is the short-term cycle-to-cycle variation of pitch, whereas the analogous quantity for amplitude is called shimmer.)