Detection of Children Abuse by Voice and Audio Classification by Short-Time Fourier Transform Machine Learning Implemented on Nvidia Edge GPU Device

Children's safety in the home has become an increasing social concern, and the purpose of this experiment is to use machine learning to detect child abuse scenarios in order to improve children's safety. This experiment uses machine learning to classify and recognize children's voices and to predict whether the current sound the child is making is crying, screaming or laughing. If a crying or screaming sound is detected from a child, an immediate alert is sent to the relevant personnel so that they can sense what the child may be experiencing and respond in a timely manner in the blind spot of the surveillance. Together with the hybrid use of video image classification, the accuracy of child abuse detection can be greatly improved. This greatly reduces the likelihood of children being violently abused in childcare centers and allows staff to stop impending or recent child abuse incidents in a timely manner. The dataset collected for this experiment was derived exclusively from sounds recorded on-site at the children's home, including cries, laughter, screams, and background noise. In this experiment, the short-time Fourier transform is used to transform the sound signals into spectrograms, and the image signals are imported into a convolutional neural network for training and prediction, and the final trained model has an accuracy of about 92% for sound detection.