Anomaly Detection in Raw Audio Using Extreme Learning Machine

Anomaly detection is a crucial problem that has garnered attention across various research areas and application domains. Numerous anomaly detection techniques have been developed, with specific focus on domains such as sound or speech recognition. Sound recognition applications are commonly employed in health system monitoring and control, making them particularly relevant in this study. The objective of this research is to detect anomalies in sound, enabling informed decision-making regarding the handling of such sounds. For this study, the TUT Rare Sound Events 2017 dataset is utilized, comprising 2987 audio files. These files encompass isolated sound events for each target class with the noise background noise from everyday acoustic scenes. The dataset is divided into a training and testing set with the split of $80:20$. The Extreme Learning Machine (ELM) method is employed for the learning process. The accuracy and performance of the ELM method are evaluated through calculations. The results reveal that the ELM method demonstrates promising capabilities in detecting anomalies within raw audio data, achieving an accuracy of 93.98%. Notably, the highest accuracy of 92.13% is achieved when detecting the anomaly of baby cry. These findings highlight the effectiveness of the ELM method in anomaly detection within sound data.