Autism spectrum disorders (ASD) represent a wide range of neurodevelopmental disorders. It is difficult to make a diagnosis before 18 months, while affected children may have special vocal characteristics as early as 6 months. Taking care of the child early could promote his cognitive development. So we developed ASDLearn, a diagnostic assistance tool for doctors using Deep Learning. It allows the voice recording of a child, generates a spectrogram image - a 2D representation of the voice signature - associated with it, and compares it to our database through a network of neurons that we have pre-trained.
Supervisor : Nadia MADAOUI