German specialists have created an artificial intelligence (AI) model which was taught to accurately recognize human emotions through voice.

The respective research was published in the journal Frontiers in Psychology.

The researchers found that it took 1.5 seconds of a recording of a model's speech to determine its mood.

The models achieved human-like accuracy, says study author Hannes Diemerling.

The model analyzed the voice and determined the mood of the given person using neural networks. Using deep neural networks (DNN), it analyzed frequency and pitch, and using convolutional neural networks (CNN), the model learned to determine emotion through the rhythm and texture of the voice. Finally, a hybrid model (C-DNN) enabled the creation of a visual spectrogram for emotion prediction.

According to Diemerling, in the end, the model accurately determined the mood of the interlocutor.

If the models outperformed humans, it could mean there could be patterns we don't recognize, the researcher noted.