Detection of Children Abuse by Voice and Audio Classification by Short-Time Fourier Transform Machine Learning

Aug 6, 2026

Upload by Logic

Children’s safety in care settings has become an urgent social concern, and this experiment applies machine learning to recognize child-abuse-related audio scenarios to enable timely intervention. The system classifies children’s sounds—crying, screaming, and laughing—and immediately sends alerts to responsible personnel when crying or screaming occurs, including through surveillance blind spots. A hybrid approach with video image classification targets improved detection accuracy. Audio recordings are converted into spectrograms via Short-Time Fourier Transform and classified using a CNN, reaching about 92% accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Abstract

Introduction

Related works

Machine Learning

Neural Networks

Short-Time Fourier Transform

The combination of STFT and CNN

Our approach

Formatting of the Datasets

Conversion of signal domain

Detection of Children Abuse by Voice and Audio Classification by Short-Time Fourier Transform Machine Learning implemented on Nvidia Edge GPU device Jiuqi Yan1 , Yingxian Chen1, W.W.T. Fok 1 1 Electrical and Electronic Engineering ,The University of Hong Kong, Hong Kong *Corresponding author’s email: yjqhku@connect.hku.hk Abstract. The safety of children in children home has become an increasing social concern, and the purpose of this experiment is to use machine learning applied to detect the scenarios of child abuse to increase the safety of children. This experiment uses machine learning to classify and recognize a child's voice and predict whether the current sound made by the child is crying, screaming or laughing. If a child is found to be crying or screaming, an alert is immediately sent to the relevant personnel so that they can perceive what the child may be experiencing ina surveillance blind spot and respond in a timely manner. Together with a hybrid use of video image classification, the accuracy of child abuse detection can be significantly increased. This greatly reduces the likelihood that a child will receive violent abuse in the nursery and allows personnel to stop an imminent or incipient child abuse incident in time. The datasets collected from this experiment is entirely from sounds recorded on site at the children home, including crying, laughing, screaming sound and background noises. These sound files are transformed into spectrograms using Short-Time Fourier Transform, and then these image data are imported into a CNN neural network for classification, and the final trained model can achieve an accuracy of about 92% for sound detection. 1. Introduction Children in school or children home are encountering risk of home accident, fighting among children and being abused. There were a few incidents that children in the children home in Hong Kong were abused by their caretaker and police investigation is required. The government called for using latest Artificial Intelligent to strengthen the monitoring and supervision of children home and generate real-time alerts to supervisors if there is any violent or abnormal behavior detected. Apart from using CCTV video image, the sound in the children home could also provide signal to reflect the situation happening in the children home. This research project develop algorithm and build alight-weight AI model for the analysis of audio wave form and classify the type of sound to assist the classification of the caretakers and children’s behavior. 2. Related works 2.1 Machine Learning Machine learning has achieved great success in the past decades in the direction of image classification12, face recognition, unmanned autonomous vehicles, speech recognition, etc3.Linnaeinma4 conceived and proposed the model of BP neural, i.e. - the inverse model of automatic differentiation model. However, it did not create an academic wave at that time, but stagnated for a decade, and the backpropagation algorithm described in detail The backpropagation algorithm described in detail was introduced by Weibos5. After a few years, there have been algorithms combined with training, and many scholars involved in the field of neural network research at that time proposed the idea of combining MLP and BP training67. 2.2 Neural Networks Neural networks are an important machine learning technique with an overall structure similar to the nerves of the human brain and are designed to implement human brain functions. The most famous convolutional neural network in deep learning was proposed by Lecun et al8. It was the first true multilayer structural learning algorithm that uses spatial relativity to reduce the number of parameters to improve training performance. Based on the original multilayer neural network, a feature learning part was added, which mimics the human brain's hierarchy on signal processing. 2.3 Short-Time Fourier Transform The Fourier transform only

FAQs

What does the proposed system detect and classify?

It classifies a child’s voice and recognizes whether the sound corresponds to crying, screaming, or laughing.

How are audio signals processed before classification?

Recorded sound files are transformed into spectrograms using Short-Time Fourier Transform, and the resulting image data are fed to a CNN.

What is the purpose of sending alerts in the workflow?

When crying or screaming is detected, an alert is immediately sent so personnel can perceive potential abuse situations and respond in a timely manner, reducing the likelihood of violent abuse.

Aug 6, 2026

Upload by Logic