This AI is able to detect a person's lack of sleep by voice

Il est possible par le simple son de la voix de déterminer si un individu est en manque de sommeil. tatyana_tomsickova/Getty Images

L'intonation de la voix peut trahir un manque évident de sommeil. Des chercheurs viennent précisément d'identifier ses principales altérations. Ces résultats ont donné naissance à une intelligence artificielle capable d'identifier des sujets privés de sommeil ou pas.

According to the CNRS, three out of ten French people suffer from a lack of sleep (with nights of less than six hours). This proportion has tripled in just ten years. The causes are multiple, including exposure to screens which not only interferes with sleep time but also delays falling asleep. However, lack of sleep can cause chronic fatigue, depression, and accidents.

In a study, published in PLoS Computatinal Biology, the Vigilance, fatigue, sleep and public health (VIFASOM) research team, in collaboration with CNRS researchers, demonstrates the possibility of detecting sleep deprivation thanks to voice recordings analyzed by artificial intelligence specially trained for this.

Detecting fatigue by voice ?

For their study, they worked with a sample of 22 women, aged thirty to fifty, allowed to sleep only three hours per night, two nights in a row, in order to identify alterations in their voices. They were thus able to identify two independent effects of sleep deprivation on the voice, namely changes in prosody (melody of the sentence, modification of the voice, variation in speech rate) and in timbre (clear voice vs. voice broken).

The researchers then trained artificial intelligence to be able to detect differences in the submitted recordings. Using machine learning, the tool can now identify differences between different voice recordings and therefore detect sleep deprivation in certain individuals. The results of this study suggest future "intelligent" detecting sleep deprivation through voice analysis. They could, for example, be used in certain sectors of activity, where a drop in vigilance can have serious consequences.

The next step will be to apply the same methodology to develop other vocal biomarkers characterizing attention disorders or, more generally, speech disorders. physiological state of a person.